# Jev News > Jev News at https://jevainews.com records decision models that return a choice, a score, or a yes/no from a fixed answer set. Jev is TypeSafe AI's System One model, announced September 15, 2026. A request sends unstructured state and typed questions. The model returns a Choice, a Score, or a Noul. It does not generate text. This site is not affiliated with TypeSafe. Figures are claims from the linked sources. When a number matters, quote the linked page. Jev is TypeSafe's model. Jev News records decision models. Official product pages live at https://typesafe.ai/ and https://docs.typesafe.ai/. Jev's weights stay with TypeSafe. Community stand-ins publish their own checkpoints, and [Repos](https://jevainews.com/repos/) lists the ones these stories cite. The catalog is [Decision models like Jev](https://jevainews.com/decision-models/). Choice picks among up to 255 named options and returns a distribution. Score returns a number on a rubric the caller defines. Noul is a yes/no and returns a probability between 0 and 1, with no separate confidence field. Several questions can share one request. TypeSafe says sampling is parallel and cites 70 to 500 ms. A job that needs a written string still uses a language model. The explainer is [What Jev returns](https://jevainews.com/news/what-is-a-system-one-model/). TypeSafe's September 15, 2026 launch post says Jev gives up string generation and can't hallucinate. The same post says schema matching is guaranteed, and that the 0% on its plot is not an empirical count. The standing page is [Decision model hallucination](https://jevainews.com/hallucination/). Cloudflare's Clef card prints RAGTruth hallucination F1 at 79.4, 35.6, and 76.5 for Clef, Clef-flash, and Jev. Perplexity's 1,500-row RAGTruth accuracy is a different test: Jev 77.27%, Qwen3.8-27B 61.53%, pplx-decider 88.80%. Jev Day 2026 is TypeSafe's October 14, 2026 release. The standing page is [Jev Day 2026](https://jevainews.com/jev-day-2026/). Date headers 08:49:11 GMT to 08:49:40 GMT October 26, POSTs with jev-latest and jev-preview named jev-1.14.0. Opened that hour, the models page says both aliases point at jev-1.14.0. On October 16, 2026, Date headers 08:19:44 GMT to 08:19:46 GMT, one Box for every vehicle on a three-car PNG returned one rectangle, n = 3, confidence 0.19, 0.17, 0.13, and type array returned HTTP 400. TypeLLM's cookbook at 14.00 s prints six vehicle boxes, including Tesla's flying Roadster. A 960 by 540 frame from Elon Musk's clip chose six at 0.59. Aviz Maeir scored Jev 1.14 Box at 82% on 180 scenes, 0.21 s, $0.15 per 1k, and ≥0.7 was right 107 of 108. Date headers 00:28:47 GMT to 00:28:48 GMT, this desk's Continue button n = 3 averaged IoU 0.93, confidence 0.91. At 00:28 UTC, TypeSafe's jev-latest still returned jev-1.13.0. Opened that hour, OpenRouter listed typesafe/jev-1.14 at $0.042 per million input tokens, output free, a 262K context, and TypeSafe P50 0.19 s. GET /api/v1/models did not list that id. This desk did not call OpenRouter. LiteLLM's October 15 post adds day-0 images, box, and long context. On October 15, Yamashita's Kyoto-ben 60 scored 0.833 on jev-1.14.0 against 0.817 on jev-1.13.0, majority vote matched on all 60. Jev4Mellea's 12 synthetic scenes on jev-1.14.0 accepted none of 6 positives at 0.10 / 0.90, coverage 5 of 12. AI BENCHY scored Jev 1.14 at 2.3 against Jev 1.13 at 2.6. SosyalDijital scored both models 29 of 30 on 30 synthetic Turkish tickets. TypeSafe's internal chart prints overall intelligence 60.0 for jev 1.14.0 and 58.9 for jev 1.13.0. The image says Please don't trust them. 60.0 is not the JevBench composite. Images need the header TypeSafe-Danger-Zone-Images: true. [Box](https://jevainews.com/news/jev-box/) is a Jev 1.14 question type. [Long context](https://jevainews.com/news/jev-long-context/) raises both caps to 262k tokens with TypeSafe-Danger-Zone-Long-Context: true. Default limits stay 64k per request and 32k per question. Open decision checkpoints start from a public language model because a closed API does not release the weights a decision head has to read. The write-up is [Open decision models](https://jevainews.com/news/open-decision-bases/). Jeff v1.3, opened October 8, 2026, prints a mean of 30.4% for untrained Qwen3.5-0.8B and 92.8% with an adapter across 13 of its own tests. Unsloth's October 7 documentation also fits Llama 3.2 3B and Gemma 4 E4B. Those rows are separate tests. Opened October 7, 2026, TypeSafe's models page lists jev-1.13.0 at $0.042 per million input tokens, free output, 100K tokens per second, and 80 requests per second. Opened October 1, 2026, TypeSafe's models page lists jev-1.13.0 at $0.042 per million input tokens, free output, 100K tokens per second, and 40 requests per second. The page says those limits can change without notice. The whole request is 64k tokens. State plus the longest question is 32k. Each host publishes its own rate. Doors this desk has cited are on [Access](https://jevainews.com/access/). Signup status is the [signups story](https://jevainews.com/news/typesafe-signups-paused/), which this desk updates in place. On October 8, 2026 the same signed-in console still showed available credits of $0.00, no payment method, and no billing history. Usage for the last 7 days was $0.00, 0 tokens, and 0 requests. On October 6, 2026, at 17:05 UTC, a new console account showed available credits of $0.00 and an empty billing history. On September 30, 2026 one POST to https://api.typesafe.ai/v1/systemone with model jev-latest returned HTTP 200 and model jev-1.13.0. That response held a Choice, a Score, and a Noul. Scores here were published by the people named on each page. This desk did not rerun them. JevBench v1.4.2.2 and v1.5.0 are different boards. Opened October 19, 2026, JevBench JSON revision v1.6.29 ranks 184 open systems and 38 hosted ones. basal-omni-1.0-ultra is first of 184 on Capability at 82.5. kettei-fat stays first of 38 hosted at composite 77.8. Jev 1.14.0 is ninth at 70.2. Jev 1.13.0 is seventh at 71.5. Opened October 17, 2026, the API board named revision v1.7.46 and 39 hosted systems, with kettei-fat first at 77.8. Opened October 11, 2026, that board ranked 27 hosted systems, Sage 1.3.0 first at 74.0. 72.1 stays the v1.5 composite. The current table and the older cell are both in the [JevBench story](https://jevainews.com/news/jevbench/). Collected measurements are on [Evals](https://jevainews.com/evals/). Similar tests sit side by side on [Compare](https://jevainews.com/compare/). Failure modes are on [Limits](https://jevainews.com/limits/). TypeSafe's Master Customer Agreement section 2.3(f) forbids customers from publishing benchmarks of the Services. [Method](https://jevainews.com/method/) records how this desk treats that rule. OpenAI Jev is this desk's name for OpenAI's Decisions API, written up at [OpenAI Jev](https://jevainews.com/news/openai-jev/). On October 23, 2026, Date headers 02:23:12 GMT and 02:23:13 GMT, two POSTs from this desk returned HTTP 200 from gpt-6-luna. OpenAI Developers posted general availability on October 22, 19:07 UTC. The guide prices uncached input at $0.10 per million tokens and cached input at $0.01, with no output charge. Requests now hold up to 400 questions. On October 6, 2026 OpenAI Developers posted it in public beta. The same day a POST to https://api.openai.com/v1/decisions with model gpt-6-luna returned HTTP 200, 420 input tokens, and 0 output tokens. The October 6 guide priced that input at $0.10 per million tokens and listed no output charge. Choice and score answers included confidence. An empty POST returned HTTP 400, missing model. October 2 and October 4 calls had returned HTTP 403. The DevDay recap, opened October 6, still said limited preview. Later the same day three texts went to Decisions and to Jev once each. A cat sentence scored 0.98 on both. A Stripe ticket from the TypeSafe quickstart was technical on both, at confidence 0.99 and 0.74. Frustration on that ticket matched at 1.0. The Decisions host is OpenAI's. The paired Jev calls went to api.typesafe.ai. On October 8, 2026 ten urgency calls from this desk had a median clock of 668 ms on Jev and 387 ms on Decisions, in separate passes. The three-question medians were 719 ms and 389 ms. Benchmark Heaven's API board, opened October 8, 2026, ranks OpenAI Decisions seventh at composite 62.5 and Jev 1.13.0 fourth at 71.5. The October 7 opening ranked Decisions fifth, interval 51.6 to 64.6, and Jev third. HiringCafe's October 7 thread prints Spearman 0.74 for Jev and 0.71 for OpenAI on query-to-job relevance, and 0.71 against 0.67 on resume-to-job. The thread does not print the pair count. Clef and Clef-flash are Cloudflare's Apache 2.0 decision models, posted October 1, 2026 and written up at [Clef](https://jevainews.com/news/cloudflare-clef/). The October 9 blog lists Clef at $0.24 and Clef-flash at $0.038 per million input tokens, and a hosted 24k window for Clef-flash. The October 1 cards listed Clef-flash at $0.09 and a 65,536-token window. The October 1 blog's median latency row is 209.3 ms and 38.8 ms against a printed Jev cell of 524.1 ms. typesafe/jev on the same host is a separate card. The catalog is [Decision models like Jev](https://jevainews.com/decision-models/). [Clef-omni](https://jevainews.com/news/clef-omni/) is Cloudflare's October 9, 2026 decision model. The blog prices input at $0.15 per million tokens. BANKING77 macro-F1 on that post's ten-row table is 94.8. A 21-second video with sound scores in about 1.5 seconds. The weights are Apache 2.0, from Qwen3-Omni-30B-A3B-Instruct. This desk did not call the API. [PurpleMIST-Flash-1.0](https://jevainews.com/news/purplemist-flash/) scores 0.668 accuracy on 15,050 translated African-language decisions. DeepSeek-V4-Pro on that harness is 0.669, and KL is 0.214 against 1.125. The card says the checkpoint is Apache 2.0, about 7.9B parameters, from Qwen3.5-9B-Base, and text only. A self-run of Decision Index edition 0.3 prints a public index of 46.7. That index is a different test from the 0.668. This desk did not run the weights. [vegaml](https://jevainews.com/news/vegaml/) scores 75.4% on 800 phishing emails against Jev 1.13.0 at 61.9%. The default checkpoint is 0.8B on frozen Qwen3.5-0.8B, Apache 2.0, with a 4B sibling. Decision 2.0's Vega, which prints Decision Index 0.2.1 at 56.5, is a different checkpoint. This desk did not run the weights. [Sage 1.3.0](https://jevainews.com/news/sage/) is Levanto's hosted decision API. Opened October 11, 2026, the JevBench API board ranks it first of 27 at composite 74.0. The pricing page lists Pro input at $0.042 per million tokens and output at $10 per million for reasoning. Answers on that page are free. This desk did not call the API. [H2O-Lightning-4B v1.1](https://jevainews.com/news/h2o-lightning-4b/) prints JevBench composite 72.5 against Jev 1.13.0 at 71.5 on H2O's October 7, 2026 blog. That reading is board revision v1.7.17 and 104 open-weight systems. The blog says the weights are Apache 2.0, from Qwen3.5-4B. This desk did not run the weights. [OpenBles Firewall 1](https://jevainews.com/news/openbles-firewall/) caught 119 of 120 agent attacks and flagged none of 120 benign look-alikes. The classic split is 55 of 60, with 2 of 56 benign prompts flagged. The adapter is MIT, 31.8 MB, on fastino/GLiNER2.5-Decide. This desk did not run the weights. [d3-ultra](https://jevainews.com/news/d3-ultra/), posted October 27, 2026, prints Decision Index 0.4.1 Full 70.18, second of 229, opened October 26. The card's private SR-Text mean is 87.2. Vision board edition 0.4 full is 70.59. Apache-2.0, from Qwen3.8-27B. Date headers 06:25:57 GMT and 06:25:58 GMT October 27, GET /v1/models listed d3-ultra, POST without a key returned HTTP 401. This desk did not run the weights. [d3-turbo](https://jevainews.com/news/d3-turbo/), posted October 20, 2026, prints Decision Index 0.4.1 Full 67.31 and vision full 72.46, second of 56, opened October 26. The free API aliases vllm-sr/d3 to d3-turbo. The card prints 255 questions about one photo at 2.12 s against d3-flash at 32.94 s. Apache-2.0. This desk did not run the weights. [Decision Model Factory](https://jevainews.com/news/decision-model-factory/), posted October 27, 2026, is Yuma Ichikawa's arXiv 2610.34956. Table 1 prints DF-JA-Large at 67.63 on S1MB Japanese Task Avg against a fixed public snapshot high of 60.05. A Hugging Face search that day returned no matching public checkpoint. This desk did not run a build. [hunch](https://jevainews.com/news/hunch/), posted October 27, 2026 by Matt Mastracci, turns EmbeddingGemma2 into a CPU /v1/systemone endpoint at published clocks of 20 ms on text and 100 ms on images. A later thread says Cloudflare Clef teaches a local student in real time, and open-source launch was planned for later that week. Opened October 28, public search found no hunch GitHub repository and no hosted POST. This desk did not call an endpoint. 20 ms is not Clef-flash's October 1 median of 38.8 ms. [OR-decision-4B](https://jevainews.com/news/or-decision-4b/), posted October 30, 2026 by alphaXiv, prints JevBench 193 of 231, 83.55% on the Hugging Face card. The launch chart prints Choice 79.7 against Jev 79.6, and sarcasm 53.6 against Jev 70.7. Apache-2.0 LoRA on Qwen3.5-4B, 117,186 train examples. Do not merge 83.55 with 79.7 or with Decision Index. This desk did not run the weights. [MiniCPM5-2B-Jev](https://jevainews.com/news/minicpm5-2b-jev/), posted October 28, 2026 by OpenBMB, prints 78.79% on JevBench, first among open models at or under 2B, and decision-v7 86.41% with ECE 0.0307. Apache-2.0 on MiniCPM5-2B. Weights ytbai/MiniCPM5-2B-Jev. The same README table prints hosted Jev 1.13.0 at 86.6%. Do not merge 78.79 with 86.41. This desk did not run the weights. [BranchOne](https://jevainews.com/news/branchone/), posted October 29, 2026 by Necati Demir, returns one probability per supplied candidate from Qwen3.5-0.8B-Base plus LoRA. His A100 Colab median on ten short text questions is 0.305 s. Apache-2.0. Weights ndemirdemir/BranchOne. No published JevBench score. 0.305 s is not Celeris 67 ms and not hunch 20/100 ms. This desk did not run the weights. [Decision 3.0](https://jevainews.com/news/decision-3-0/) is six Apache-2.0 checkpoints posted October 11, 2026. Opened October 26, Decision Index 0.4.1 Full score for d3 is 67.15. The October 16 card reading of edition 0.3.2 was 66.3. The free API aliases vllm-sr/d3 to d3-turbo. The October 11 reading was 64.7 on edition 0.3.1. The highlight prints public-suite 65.45 for edition 0.3. This desk did not run the weights. [Mizan 4B](https://jevainews.com/news/mizan-4b/) averages 78.7 on 3,000 real Persian records, against the card's Jev cell of 83.7. The Q4 GGUF mean is 78.3. The card says the weights are Apache-2.0, from Qwen3.5-4B, and text only. This desk did not run the weights. [THX-01](https://jevainews.com/news/thx-01/) averages 98.4 on 2,843 support tickets, against Jev 1.13 at 97.4 on the same four sets. Number lookup on a different set is 93.4, and Jev on that line is 96.6. The card says 322M parameters, Apache 2.0, on mmBERT-base. The October 11 post calls the API free. The docs page does not print a price. This desk did not call the API. [Quyet-1.0-Large](https://jevainews.com/news/quyet/) prints frozen JevBench v1.6.0 at official score 71.39, rank 1 of 92, and Capability 81.7. The October 8 open-weights rank line still says Capability 81.7 and official composite 3. The card says Apache-2.0, 31.3B, from Gemma 4 31B. This desk did not run the weights. [Blink v0.3](https://jevainews.com/news/blink/) prints a self-run of Decision Index 0.2.1 at 57.48, against the board's Jev cell of 57.91. The card's license field is Apache-2.0. The base is Gemma 4 26B-A4B. This desk did not run the weights. [SubQ Snap](https://jevainews.com/news/subq-snap/) is Subquadratic's hosted decision model, posted October 12, 2026. On October 23, 2026, Date headers 02:23:17 GMT to 02:23:20 GMT, three POSTs from this desk returned HTTP 200. Docs name subq-snap-1.1 at $0.04 per million input tokens. GET /v1/models listed subq-snap-1.1 and subq-snap-preview. Artificial Analysis Decision Index prints Snap 1.1 at 65, tied with d1-ultra and Elephant 9B. The October 12 blog prints Decision Index 0.2.1 at 61.7 against a Jev cell of 57.9. Docs opened October 12 priced input at $0.05 per million tokens and said output is free. The call is POST https://api.subq.ai/v1/systemone. The October 12 model id was subq-snap-preview. The window is 384K tokens of text and images. On October 13, 2026, at about 03:00 UTC, two POSTs from this desk returned HTTP 200. Urgency noul was 0.98, 119 input tokens, 0 output, desk clock 1,195 ms. Three questions chose technical at confidence 0.90, score 0.98, noul 0.98, 235 input tokens, desk clock 2,291 ms. GET /v1/models with the same key listed that one id. An earlier empty POST returned HTTP 422, and a POST with no key returned HTTP 401. Opened October 13, 2026 UTC, the Decision Index space headers v0.3.2, 149 models, Jev 60.1, and a search found no Snap row. Drex v1.5 weights, posted October 9, 2026 and written on the [Drex 1.5](https://jevainews.com/news/drex-1-5/) page, print Decision Index 0.3.1 public index 58.08 against Jev 1.13.0 at 57.96. The license is Nace.AI Open RAIL-M. Commercial use above $1,000,000 of revenue or funding needs a separate license. That 58.08 is not the September 30 chart's 58.28, and it is not Drex DLM's 52.31. This desk did not run the weights. [llamacpp-jev](https://jevainews.com/news/llamacpp-jev/) puts /v1/systemone in front of llama-server. llama.cpp v0.6.0, published October 5, 2026, serves that path for the October 2 decision GGUFs, for Bespoke-Nimble-9B-v3, and for Clef. The v3 card prints Decision Index 0.2.1 at 56.88 under CC BY-NC 4.0. That figure is not the 324-row Nimble holdout. The server README at that tag names Clef as an image model. The October 2 one-question medians on one RTX PRO 6000 stay 3 ms, 5 ms, 12 ms, 36 ms, and 43 ms. Opened October 6, the kishida branch page still says the endpoint is not upstream. [Databricks ai_decide](https://jevainews.com/news/databricks-ai-decide/) is a Beta function, posted October 1, 2026. REST is POST /api/2.0/ai-functions/ai-decide. The docs do not name the model and do not print a price. [NIRNAY 450M](https://jevainews.com/news/nirnay-450m/) prints 0.8792 on 3,080 Banking77 rows, and the card says the Jev cell was not measured in that repo. [WaterSheep](https://jevainews.com/news/watersheep/) serves POST /v1/systemone on port 8766 and prints 77.8% in distribution and 61.2% on held-out datasets, with no Jev column. [StartLux-Decision](https://jevainews.com/news/startlux-decision/) v1.1 27B Instant prints Full score 69.12, first of 174 on Decision Index 0.3.2. The October 3 self-run of edition 0.2.1 stays 63.88. Weights are CC BY-NC 4.0. [Drex Max](https://jevainews.com/news/drex-max/), posted October 19, 2026, prints Decision Index 0.3.2 public 67.59 against Jev 1.14 at 60.70. Docs price input at $0.06 per million tokens. Date headers 08:50:01 GMT and 08:50:02 GMT October 20, GET /v1/models with no key returned HTTP 401. Date header 02:23:25 GMT October 23, GET of OpenRouter id nace/drex-max still returned HTTP 404. [Sieve-31B](https://jevainews.com/news/sieve-31b/) prints Full score 63.8 on Decision Index 0.3.2. [Clairic](https://jevainews.com/news/clairic/) is Cloudflare's System One extraction model, posted October 19, at $0.10 per million input tokens. The blog cascade is 19.3 s against 61.7 s. [Datadog](https://jevainews.com/news/datadog-jev/) prints Jev Noul recall 82.90% to 3.88%. Date headers 08:50:46 GMT and 08:50:47 GMT October 20, this desk's jev-1.13.0 is_risky was 0.94 then 0.13. [basal-omni-1.0-ultra](https://jevainews.com/news/basal-omni/) is Remek Kinas's Apache-2.0 Gemma 4 31B checkpoint. Opened October 19, 2026, JevBench ranks it first of 184 open systems on Capability at 82.5. The card prints Werdykt v1 at 0.879. This desk did not run the weights. [basal-1.5](https://jevainews.com/news/basal-1-5/) prints 72.12% on Werdykt v1, against Jev 1.13.0 at 81.64%. [jevmod](https://jevainews.com/news/jev-nsfw/) scores sexual labels on OpenAI's moderation eval at AUROC 0.984, F1 0.866 at 0.80, remeasured September 23, 2026. ZeroNSFWBot, in that same write-up, scores pictures on the host and asks Jev about text when a key is set. [Tev1](https://jevainews.com/news/tev1/) is Together AI's experimental Qwen3.5 decision model. The September 23, 2026 post prices serverless input at $0.042 per million tokens. Opened October 7, 2026, the model page lists $0.04 per million input tokens and free output, and the chat call returns one letter. Ollama's library page serves tev1 and tev1:0.8b at /v1/systemone. Its mean on 3,880 decisions is 73.3% and 63.5%, against Nimble at 75.7% and a printed Jev cell of 76.0%. The Hugging Face card says the fine-tune license is still being finalized. [Mercury Decide](https://jevainews.com/news/mercury-decide/) is Inception's hosted call. Opened October 8, 2026, the JevBench API board ranks it third at composite 72.4. The free model page returned the same day, priced Free, with up to 14 decisions a second and no JevBench score. The September 30 post cites up to 14 decisions a second and prints no score. [Microsoft-Decision-1](https://jevainews.com/news/microsoft-decision-1/) is Microsoft's hosted decision model, posted October 9, 2026. Command Line prices input at $0.042 per million tokens and says output is free. The post says the model was 4.5 times quicker than Quyet-1.0-Large and 35 times quicker than GPT-6 Sol on Microsoft's own comparison, and it does not print an accuracy. The Foundry catalog, opened October 10, lists a 32,768-token window, says the weights are not distributed, and marks the lifecycle Generally available. Opened October 10, JevBench's API page still said release v1.6.1 under board v1.7.29, and a search of that text found no Microsoft-Decision-1. Opened October 11, 2026, the same API board ranks Microsoft-Decision-1 sixth of 27 at composite 69.1. The v1.7.31 note prints Capability 70.78, Calibration 84.40, and Composite A 69.11. This desk did not call the API. [TypeLLM](https://jevainews.com/news/typellm/) constrains an autoregressive model with a JSON Schema. Its September 23, 2026 file scores Qwen3.8-27B at 195 of 231 public JevBench tasks without thinking and 228 of 231 with thinking. The hard tier in that file moves from 76 of 111 to 109 of 111, and easy falls from 48 of 48 to 47 of 48. The homepage, opened October 10, prices input at $0.05 per million tokens and thinking at $0.50. The hosted name is typellm-latest. The 228 of 231 run is RadixArk/Qwen3.8-27B-NVFP4-BF16-LMHead, not that hosted name. On October 16, 2026, at 05:44 UTC, TypeLLM posted that Jev 1.14 cannot return an array of vehicle boxes, including Tesla's flying Roadster. The cookbook at 14.00 s prints six boxes. This desk did not call the API. Strands Decider 2B is an Apache 2.0 local checkpoint, posted October 1, 2026 and written up at [Strands Decider](https://jevainews.com/news/strands-decider/). v19 scores 167 of 231 on the public JevBench tasks, median 115 ms on an RTX 3090. That rank is the September 25 v1.4.2 board, 89 systems. [HydroJEV](https://jevainews.com/news/hydrojev/) reports jev-1.13.0 at mean macro-F1 0.63 on simulated in-distribution alarms. An agreement gate settles 69 of 200 and 76 of 200 fresh windows, and 32 of the 76 look-alike accepts were wrong. [DimABSA](https://jevainews.com/news/jev-dimabsa/) prints Task 1 RMSE 1.0645 from frozen jev-1.13.0 and 488 fitted coefficients. [jev-doc-search](https://jevainews.com/news/jev-doc-search/) marks both of two filing questions correct. [WrangleBench](https://jevainews.com/news/wrangle-bench/) prints 137.3 perfect matches per 200 with Jev and 105.1 without. [MMAI Gym](https://jevainews.com/news/mmai-cyp/) is 12,563 molecules: a gradient-boosting ensemble above Jev 1.13 on five CYP assays, with Jev's probabilities stopping at 0.69. [You.com's risk sweeps](https://jevainews.com/news/risk-analysis-server/) cost $0.43 for three escalated runs, and a blind read ties at 4.25. [Bud Decision Studio](https://jevainews.com/news/bud-decision-studio/) serves eleven local models at port 8420. [DSPy 3.4.0](https://jevainews.com/news/dspy-jev/) calls Jev from Predict. A patterns repo pins jev-1.13.0, and a shell gate refuses below 0.2 and runs at 0.6 or above. [GLiDE](https://jevainews.com/news/glide/) v1.2, posted October 22, 2026, prints Decision Index 0.4.1 Full 71.5, rank 1 of 229, opened October 26. Docs price input at $0.04 per million tokens. Date headers 08:49:09 GMT to 08:49:11 GMT, GET /v1/models omitted GLiDE, POST /v1/systemone HTTP 401. JevBench lists Fastino GLiDE (hosted API, 22 Oct 2026) at composite 76.4, rank 3. The October 1 self-run of Decision Index 0.2.1 stays 64.81 against the board's Jev cell of 57.91. Decision Index 0.3 listed a GLiDE no-thinking row at 60.2. The October 8 pricing reading was $0.15. The earlier docs reading was $0.30. [Jeva](https://jevainews.com/news/jeva/) prints 0 flips on 600 Choice items and BANKING77 at K=16 91.62 against Jev-1.13.0 at 82.50. Apache-2.0, Qwen3.5-4B. This desk did not run the weights. [Abe's 1,774 posts](https://jevainews.com/news/snowflake-ai-filter/) cost about $0.055 on Jev Noul against about $1.87 on Snowflake AI_FILTER. AUC 0.9685. Date headers 08:49:12 GMT to 08:49:40 GMT October 26, this desk's jev-1.14.0 noul on his Lightdash example was 0.04. [GEV-26B-Decide](https://jevainews.com/news/gev-26b-decide/) prints an adaptive 62.48 on another self-run. Adaptive thinking on Knowledge misses the kit's latency cap. Neither score is a row in the September 28 file. Fastino's October 3 chart prints Clef at 61.2 and Clef-flash at 57.1, captioned as self-reported, and pplx-decider at 56.39. The JevBench API board opened October 8 prints GLiDE at composite 30.6, rank 14, at $0.137 per 1,000 decisions, and capability 77.0 outside the cost cap. The October 7 opening listed that row at rank 11. [Decision 2.0](https://jevainews.com/news/decision-2-0/) is six Apache-2.0 checkpoints posted October 3, 2026. Vega's card prints 56.5 on its own Decision Index 0.2.1 run and does not print a Jev row. Decision 3.0's d3 card, a later release, prints 66.3 on edition 0.3.2, opened October 16, and 64.7 on the October 11 reading of edition 0.3.1, and does not replace that 56.5. [Vela 2.0](https://jevainews.com/news/vela-2-0/) is four Apache-2.0 checkpoints posted October 7, 2026. The 9B card prints Decision Index 0.2.1 at 41.63 with Noul calibration on, and 41.09 with it off. The evaluation file prints the Lux base at 46.23 and does not print a Jev row. The 0.3B tokenizer carries the Gemma Terms of Use. [fsdecide](https://jevainews.com/news/fsdecide/) scores 91.3% of 80 review items against jev-latest at 56.3%, and 60.5% against 90.5% on the authors' general set. [JEV-27B-VL](https://jevainews.com/news/jev-27b-vl/) finishes 15 of 20 pick-and-place scenes. Its six-benchmark mean is 84.07 against a hosted Jev cell of 83.85, and that JevBench column is not the official board. [Open d1](https://jevainews.com/news/open-d1/) is Liquid AI's October 7, 2026 pair of open checkpoints, d1-3B and d1-omni-600M. The d1-3B card prints Decision Index 0.2.1 at 48.57, scored with the official scorer and not submitted to the board. Winnow-12B on that table is 50.02. d1-omni-600M prints 15.95. The hosted chart at 58.9 and the JevBench composite 73.0 are other tests. The d1-3B LICENSE file is LFM Open License v1.0, with commercial use conditioned on annual revenue under $10,000,000. [pplx-decider](https://jevainews.com/news/pplx-decider/) posted v1.2 on October 16, 2026. The card prints Decision Index 0.3 overall 68.09 on joint requests against v1.1 at 62.30. Docs price v1.2 at $0.02 per million input tokens and flash at $0.01. GET of the OpenRouter ids returned HTTP 404. The October 1 panel still prints 85.71% against jev-1.13.0 at 84.51% and Qwen3.8-27B at 74.76%. Jev is ahead on six of the eleven named sets. Opened October 7, 2026, Decision Index 0.3 ranked v1.1 first at 62.8, with Jev 1.13.0 as a reference at 60.1. The v1.1 card's own table prints 61.56 against Jev at 57.9. [Drex DLM](https://jevainews.com/news/drex-dlm/) is the open checkpoint Nace posted on October 6, 2026 under the name Drex 1.1. The card scores Decision Index 0.2 at 52.31 against the kit's Jev cell of 51.67. Weights are CC BY-NC 4.0. The call is POST /v1/systemone. [PolicyLM-1.7B](https://jevainews.com/news/policylm-1-7b/) returns a score from 0 to 1 for each category in a policy, up to 16 in one pass. The card prints custom-policy accuracy 0.842 at a 0.5 cutoff, and the table has no Jev column. [StreamDecisionBench](https://jevainews.com/news/stream-decision-bench/) prints Kev-27B at 62.15 and the Jev API at 59.63, normalized log-AUC over 0.5 to 8 seconds. [CUA Speedrun](https://jevainews.com/news/cua-speedrun/) prints Jev at 9.8% and 125 seconds a task on 50 OSWorld tasks. [A 37-dataset abstract](https://jevainews.com/news/jev-benchmarking/) counts 346,009 requests for under $10 and says Jev beats Qwen3.8-27B on 27 of 37 datasets. An October 2 pilot on released Open-Jev-2B moves a synthetic test from 84 of 128 to 99 of 128. [Jev Router](https://jevainews.com/news/jev-router/) is OpenRouter's chat model id typesafe/jev-router. Arena's October 7, 2026 thread prints +3.6% net and a median $0.155 a task on more than 4,700 sessions. Against DeepSeek V4.1 Flash (Max), the post says similar performance costs 38% more and the median request latency is 1.7 times higher. The model page FAQ, opened October 8, says the router has no price of its own. This desk did not call it. [Step-Jev](https://jevainews.com/news/step-jev/) is a pair of Qwen3.5-4B-Base LoRAs trained on a search agent's own traces. On October 9, 2026 this desk sent the states frozen in the repository's tests to api.typesafe.ai. jev-latest returned jev-1.13.0. The no-step state was incorrect at probability 1.0. The five-step fixture was correct at 0.05. A search whose observation is the word reply was neutral at 0.79. On a Moon-landing question written here, a no-search answer of 1965 was incorrect at 1.0, a sourced 1969 was correct at 1.0, and a repeated NASA search stayed positive at 0.53. real_jev_gate.py reads rollout files that are not in the repository, so the note's hosted AUC figures were not recomputed. On one BrowseComp-Plus seed, grading the final answer and every step scores 19.6%. Grading only the final answer scores 5.4%. No training is 15.6%. A string-match answer key is 22.9%. Hosted Jev, zero-shot, scores 0% on that page. The code is bandr-ai/bandits. The study's adapters are on a Modal volume, not a public checkpoint this desk found. [Jev RAG](https://jevainews.com/news/jev-rag/) is the name used for a typed stage after retrieval. TypeSafe's CLERC cookbook, jev-1.12, moves top-1 from 5% to 18% on 40 queries when a BM25 shortlist of 30 is reranked, in 1,200 calls at a printed $0.0645. Hindsight's listwise rerank prints LoCoMo recall@1 0.950 against a local MiniLM at 0.800. aifabrice/jev-rag prints nDCG@10 0.451 on its best NFCorpus row. This desk did not rerun them. [Bag of decisions](https://jevainews.com/news/bag-of-decisions/) is Doug Turnbull's October 8, 2026 product-search run. The blog prints WANDS mean NDCG@10 of 0.610 when GPT-5 writes the Jev questions and 0.574 for one question. Both medians print 0.561, and the CSV stores one shared float, 0.5609447680982702. The score adds 10 times the summed Nouls to retrieval, on the top 100, model jev/jev-latest. This desk checked the CSV and did not rerun the searches. [Pairwise Choice](https://jevainews.com/news/choice-bradley-terry/) is Nathan LeClaire's October 12, 2026 ranking tip, plus two eight-passage lists this desk sent to jev-1.13.0. On the Stripe shortlist a 0-9 Score put Roll secret at 8.86 and the signing-secret page at 8.07. The other six scored 2.42 or below. Pairwise Choice ranked rotating the API key third. The majority graph cycled A, F, E. Four calls used 9,664 input tokens, $0.000406 at list price. These texts are not a retrieval corpus. [JevEval](https://jevainews.com/news/jeveval/) is DeepEval's custom metric that sends Noul, Score, and Choice questions to Jev in one request and returns a weighted mean. Apache-2.0. PyPI deepeval 4.2.8 uploaded October 2, 2026. On October 14, 2026, at 01:58 UTC, two POSTs from this desk to jev-1.13.0 scored DeepEval's Tool Faithfulness example at 0.254 when the answer invented a breeze, and 0.901 when it did not. The docs walkthrough that prints 0.461 is illustrative. This desk did not install deepeval. LangChain's 500/500 and the cascade paper's 1.7 points are other tests. [TrueForge](https://jevainews.com/news/trueforge-jev/) is TrueFoundry's MIT agent harness. Akshay Pachaar's October 13, 2026 recap, and the October 12 article, put Jev on a weekly client follow-up agent that matches requests to GitHub issues. On 40 requests both arms scored 37 of 40. Printed decision costs are $0.0015 with Jev and $0.005 to $0.015 without it. Review caught 5 wrong Jev matches. On October 13, at 09:21 UTC, he wrote that 40 is too small to establish how reliably confidence routes errors to review. The 150 times line is a separate Opus rate-card estimate. npm @truefoundry/trueforge 0.178.0 requires Node 22.14 or newer. On October 14, 2026, at 02:34 UTC, two POSTs from this desk to jev-1.13.0 scored a written shortlist at confidence 1.0 and a trap shortlist at 0.51. This desk did not install TrueForge. [d1-ultra](https://jevainews.com/news/d1-ultra/) is Liquid AI's hosted decision model, posted October 15, 2026. Vals AI, October 22, 2026, 23:57 UTC, printed it as the only decision model to get all 1,960 ClaimBench claims right. Opened October 16, Artificial Analysis Decision Index prints 65 against Jev 1.14 at 57. That day's Artificial Analysis thread prints an 85% non-hallucination rate on AA-Omniscience-Decision against 2% for hosted d1, and accuracy 32% against 41%. The blog prices input at $0.042 per million tokens, output free. Model id d1-ultra. POST https://api.liquid.ai/decisions/v1/systemone. Vercel lists a 32,768-token window. Liquid's 112-evaluation average of 76.7 is a different table. Hosted d1 at 58.9 and open d1-3B at 48.57 stay on their own pages. This desk did not call the API. [mammoth-1](https://jevainews.com/news/mammoth-1/) is Base Labs' 27B checkpoint, posted October 21, 2026. The card prints Decision Index 0.4 Full 71.24, first of 208. Opened October 23, the live space is edition 0.4.1, 229 models, mammoth-1 Full 68.4 rank 3. elephant-1 is the 9B sibling, card 62.65. MIT, from Qwen3.5-27B and Qwen3.5-9B. Hosted POST https://inference.baseten.co/v1/systemone, model decision-models/baseten-mammoth-27b-latest. Date headers 02:23:21 GMT to 02:23:23 GMT, GET /v1/models with no key returned HTTP 401. This desk did not send a billed POST. 71.24 is the card. 68.4 is the live 0.4.1 bar. [Celeris-1 Decision](https://jevainews.com/news/celeris-1-decision/) is Celeris's hosted decision model, posted October 8, 2026. The blog prints 80.0% on 3,210 jev-bench items at a 67 ms median, and $0.04 per million input tokens, output free. Model id celeris-1-decision. POST https://inference.celeris.ai/celeris-1-decision/v1/systemone. Opened October 16, Artificial Analysis Decision Index lists it at 52. GET /v1/models with no key returned HTTP 401. This desk did not send a billed POST. 80.0% is not Florian's JevBench composite. [Span-02](https://jevainews.com/news/span-02/) is Respan's decision checkpoint, posted October 15, 2026. Hendrix Liu posted October 16, 23:16 UTC, that it had more usage in under 24 hours than Span-01 did in 20 days. The blog prints Decision Index 0.3 thinking off at 61.2 against Jev 1.13.0 at 58.0. The Hugging Face card prints text index 63.62 on 140,178 requests. Weights are a Gemma 4 26B-A4B fine-tune. Hosted POST https://api.respan.ai/api/v1/decisions, input_cost 0.01 on the public models list opened October 16. This desk did not run the weights. [OpenRouter Armor](https://jevainews.com/news/openrouter-armor/) is a beta plugin, posted October 15, 2026, that scores each tool call with a decision model before the call runs. The 492-call chart prints Jev 1.13 at 90.7%, 446 of 492, $0.0021 a call. On October 16, 2026, at 01:18 UTC, two POSTs from this desk to jev-1.13.0 scored git push origin main at noul 0.98 and ls ./reports at 0.01. Those two calls are not the 492. Armor is not a catalog row. [kettei-fat](https://jevainews.com/news/kettei-fat/) is Sqwish Labs' hosted decision model. Opened October 19, 2026, JevBench ranks it first of 38 hosted systems at composite 77.8, capability 86.3. Opened October 17, the API board named 39 systems at that same 77.8. ImageJevBench v0.3.0 capability is 75.9. Docs price input at $0.06 per million tokens, output free. POST https://api.sqwish.ai/v1/decide. GET /v1/models with no key returned HTTP 401. This desk did not send a billed POST. Federica's post printed #1 of 37 and intelligence 80 against 70. The blog's 93.1% on 231 public items is a different print. [HiddenLayer](https://jevainews.com/news/hiddenlayer-jev/) published a checkout test on October 14, 2026. TypeSafe Jev on jev-ultrafast gave Confirm Price 91 percent and Place order 9 percent after an ad. Date headers 07:24:57 GMT and 07:24:58 GMT October 19, two POSTs from this desk to jev-1.13.0 still chose place_order, 1.0 then 0.89 with confirm_price at 0.11. Those two calls are not the 91 / 9. Decision hijacking 1.8% to 3.5% is a different paper. [Qwen-Audio-Jev](https://jevainews.com/news/qwen-audio-jev/) is arXiv 2610.21614, submitted October 15, 2026. Choice 78.70% on MMAU-mini, about 20x against Qwen-Audio-3.1-ASR-Flash-Next. This desk did not run the model. [ToddlerJev](https://jevainews.com/news/toddlerjev/), posted October 18, prints 68.70% on MMAU-mini. Weights are TeenageWorks/Toddler-Jev under the Qwen Research License. This desk did not run them. Vercel's October 2026 AI Gateway Production Index, published October 15, prints Jev at 35.3% of paid teams in its first two weeks, against Sonnet 5.5 at 18.1%. TechCrunch, October 16, 15:33 UTC, rounds that to 35%. The September 18 post still prints nearly 13% in the first 24 hours. The write-up is [Gateway adoption](https://jevainews.com/news/vercel-gateway-adoption/). Databricks posted October 16 that TypeSafe's Jev-1.14 is available through ai_classify and ai_decide. The write-up is [ai_decide](https://jevainews.com/news/databricks-ai-decide/). TypeSafe's confidence page, opened October 3, 2026, prints Choice confidence as (p_max - 1/n) / (1 - 1/n). Score confidence uses the spread around the chosen level. Noul still has no separate confidence field. The page's 0.5 and 0.9 thresholds are an example. The write-up is [What Jev returns](https://jevainews.com/news/what-is-a-system-one-model/). The full text of every story is [llms-full.txt](https://jevainews.com/llms-full.txt). Each story link ends with the date that page was last updated. The Optional section is the older archive. ## What Jev is - [FAQ](https://jevainews.com/faq/): What a call returns, where to send one, the published price, who measured what, whether the weights are public, and what this site is. - [What Jev is](https://jevainews.com/jev/): Standing page for what Jev returns, how that differs from a language model, the three primitives, and which speed figures are TypeSafe's own unreproduced test. - [Jev Day 2026](https://jevainews.com/jev-day-2026/): October 14, 2026: jev-1.14.0 as jev-preview, opt-in images, a Box type, and 262k context. On October 16, at 08:19 UTC, one Box for every vehicle returned one rectangle, n = 3, and type array returned HTTP 400. TypeLLM's cookbook prints six vehicle boxes, including Tesla's flying Roadster. - [On October 26 TypeSafe's jev-latest named jev-1.14.0, and both aliases point at 1.14.0](https://jevainews.com/news/jev-1-14/): Date headers 08:49:11 GMT to 08:49:40 GMT October 26, POSTs with jev-latest and jev-preview named jev-1.14.0. Docs say both aliases point at jev-1.14.0, and that there is no preview build. A named jev-1.13.0 POST still named 1.13.0. At 00:28 UTC October 16, jev-latest still returned jev-1.13.0. Updated 2026-10-26. - [A green Continue button scored noul 0.99 on jev-1.14.0, n = 3, and an empty field scored 0.02](https://jevainews.com/news/jev-images/): On October 16, 2026, at 00:28 UTC, a green Continue button scored noul 0.99 on jev-1.14.0, n = 3, and an empty field scored 0.02. Aviz Maeir scored Jev 1.14 Box at 82% on 180 scenes. Jev4Mellea's 12 scenes covered 5 of 12 at 0.10 / 0.90. Updated 2026-10-16. - [TypeLLM said Jev 1.14 cannot return every vehicle box, including Tesla's flying Roadster; this desk's every-vehicle Box was one rectangle, n = 3, and type array returned HTTP 400](https://jevainews.com/news/jev-box/): On October 16, 2026, at 08:19 UTC, Jev 1.14 returned one box for every vehicle, n = 3, confidence 0.19, 0.17, 0.13, and type array returned HTTP 400. TypeLLM's cookbook at 14.00 s prints six vehicle boxes, including Tesla's flying Roadster. Updated 2026-10-16. - [TypeSafe-Danger-Zone-Long-Context raises both Jev context caps to 262k tokens](https://jevainews.com/news/jev-long-context/): The header TypeSafe-Danger-Zone-Long-Context: true raises both context caps to 262k tokens on Jev 1.14 and Jev 1.13. Default limits stay 64k per request and 32k per question. On October 15, 2026, at 01:22 UTC, this desk's header calls returned HTTP 200 on 2,122 input tokens. Updated 2026-10-16. - [Jev API](https://jevainews.com/jev/api/): POST /v1/systemone, where a console API key comes from, and the one call this desk sent on September 30, 2026. - [Jev pricing](https://jevainews.com/jev/pricing/): TypeSafe's listed input price, free output, 80 requests per second on the models page opened October 7, 2026, and the rates other hosts publish. - [Jev vs Laya](https://jevainews.com/jev/vs-laya/): Laya's Apache 2.0 weights beside the tables its authors published. Those Jev cells were not measured in the Laya repo. - [Jev benchmark](https://jevainews.com/jev/benchmark/): Opened October 19, 2026, JevBench ranks 38 hosted systems. Jev 1.13.0 is seventh at composite 71.5. Jev 1.14.0 is ninth at 70.2. kettei-fat is first at 77.8. v1.5 printed 72.1. TypeSafe's workflow multiples are a separate vendor test and were not reproduced here. - [Jev in Python](https://jevainews.com/jev/python/): typesafe-sdk, langchain-typesafe, and DSPy's TypeSafe client. Jev's weights are not a pip package. - [Decision models like Jev](https://jevainews.com/decision-models/): Hosted calls and open checkpoints that return a choice, a score, or a yes/no, including OR-decision-4B, MiniCPM5-2B-Jev, BranchOne, d3-ultra, d3-turbo, GLiDE v1.2, Jeva-4B, OpenAI's Decisions API, Cloudflare Clef, Liquid d1-ultra, Celeris-1 Decision, Span-02, kettei-fat, Drex Max, StartLux-Decision v1.1, Sieve-31B, basal-omni-1.0-ultra, ToddlerJev, pplx-decider-v1.2, and Databricks ai_decide. - [Where to start](https://jevainews.com/start/): A reading order: what Jev returns, where it fails, where to call it, and who measured what. - [What Jev returns: System One in plain language](https://jevainews.com/news/what-is-a-system-one-model/): TypeSafe defines a System One model as parallel, typed, and calibrated decisions without text generation, distinguishing it from language models operating in JSON mode. On October 3, 2026 the confidence page prints the Choice and Score formulas. Noul still has no separate confidence field. Updated 2026-10-03. - [TypeSafe leaves stealth and ships Jev, a model that refuses to chat](https://jevainews.com/news/typesafe-ships-jev/): Former OpenAI researcher Diogo Almeida announced Jev, a System One model built for typed software decisions rather than text generation, following two years in stealth. On September 30, 2026 this desk scored 32 labeled tickets. jev-1.13.0 and gpt-6-luna were each right on 32 of 32. p50 wall time was 913.5 ms and 2,698.5 ms. Updated 2026-09-30. - [TypeSafe lists nine ways Jev 1.13 is still jagged, and asks for more](https://jevainews.com/news/jev-1-13-jaggedness/): Cofounder Sasha Sheng posted that Jev is still too slow, too expensive, and too dumb, and pointed at TypeSafe's jaggedness page for jev-1.13. The docs, last reviewed September 17, name nine failure modes and ask people to file more in Discord. Updated 2026-09-30. - [Patterns](https://jevainews.com/patterns/): Recurring loops from the stories: rank a legal set, judge documents, override only when confident, route the expensive call, and recreate the model locally. - [About](https://jevainews.com/about/): What this site records, and how a citation is attached to a claim. - [Method](https://jevainews.com/method/): How a story gets on the desk, including TypeSafe's rule on customers publishing benchmarks. ## Where to call Jev - [Access](https://jevainews.com/access/): Public doors this desk has cited: TypeSafe's API, Vercel AI Gateway, Cloudflare Workers AI, Netlify, OpenRouter, Venice, and LiteLLM. - [TypeSafe reopens Jev signups and drops the free credit for new accounts](https://jevainews.com/news/typesafe-signups-paused/): On October 8, 2026 the same signed-in TypeSafe console still showed $0.00 in available credits, with no payment method and no billing history. Usage for the last 7 days was $0.00, 0 tokens, and 0 requests. On October 6, 2026, at 17:05 UTC, this desk signed in to a new TypeSafe console account that showed $0.00 in available credits. Billing history was empty. TypeSafe had paused new signups on September 22 and said on September 24 that they were still closed. On September 27 at 22:30 UTC the same account posted that signups were open, and that new signups no longer get free credits. Updated 2026-10-08. - [OpenRouter, Cloudflare, and Netlify start serving Jev](https://jevainews.com/news/jev-on-gateways/): Opened October 19, 2026, Cloudflare lists typesafe/jev-1.14.0 at $0.042 per million input tokens, output free, and a 32,000-token window. OpenRouter's typesafe/jev-1.14 listing stays 262K. GET /api/v1/models did not list that OpenRouter id. This desk did not call either host. Updated 2026-10-19. - [Venice lists Jev on its API: same key, no copy of the state](https://jevainews.com/news/jev-on-venice/): Venice put Jev on a beta Decisions API the night of September 18. Calls go to POST /api/v1/decisions with model jev-latest, billed at TypeSafe's $0.042 per million input tokens with free output. Venice says it keeps no copy of the state or the answers, and TypeSafe quoted the launch. Updated 2026-09-30. - [LiteLLM adds day-0 Jev 1.14 images, box questions, and long context on both decision routes](https://jevainews.com/news/jev-on-litellm/): LiteLLM's October 15, 2026 post adds day-0 support for Jev 1.14 images, box questions, and long context on /v1/systemone and /v1/decisions. The proxy forwards TypeSafe-Danger-Zone- headers to TypeSafe only. This desk did not run the proxy. Updated 2026-10-16. - [MotherDuck's prompt_jev() labels 100,000 AG News rows in 40 seconds for $0.50](https://jevainews.com/news/motherduck-prompt-jev/): MotherDuck shipped prompt_jev(), a SQL function that calls TypeSafe's Jev. On 100,000 articles drawn from the AG News training split, the published table is 2,484 rows per second, 89% accuracy, $0.50, and 40 seconds. GPT-5.6 Terra on the same sample is 52 rows per second, 88%, $37.58, and 31 minutes 59 seconds. Updated 2026-09-30. - [A native DuckDB extension batches Jev over SQL rows at about 2,300 rows per second on a live run](https://jevainews.com/news/duckdb-jev/): Prasanth J posted duckdb-jev, an Apache 2.0 C++ extension that asks TypeSafe's Jev from SQL. A live Choice run classified 2,049 synthetic rows in 0.887 seconds. A cached replay of the same query finished in 28.6 ms with zero API calls. Row text leaves DuckDB. Updated 2026-10-03. - [Zachi's pg-jev judges 2,000 Postgres rows in about 3.5 seconds](https://jevainews.com/news/jev-pg/): Zachi's pg-jev judges a 2,000-row Postgres table in about 3.5 seconds for about $0.012. Version 0.2.0 packs 20 rows per request. The page at pgjev.com still says 40. Jose Mejias's pg-jev is a different repository. Updated 2026-10-03. - [Jevflake puts Jev in Snowflake: noul, choice, and score as SQL functions](https://jevainews.com/news/jevflake-snowflake/): Luke Kranz published Jevflake, a dbt package (and a Terraform module) that lets Snowflake call TypeSafe's Jev API from SQL. Answers are stored once, reused, and tested like any other model. Version 0.1, run against one live account. Row content leaves Snowflake. Updated 2026-10-26. - [Vercel's form router keeps Jev's destination when confidence is at least 0.95](https://jevainews.com/news/vercel-form-router/): Vercel published a Next.js template that asks Jev to pick a team for a form submission and sends the same submission to GPT-5.6 Luna Fast when confidence is below 0.95, missing, or the call fails. The fallback does not see Jev's answer. Tests mock both models. Updated 2026-09-26. - [jevctl puts Jev on the command line, with a compaction hook that keeps lines verbatim](https://jevainews.com/news/jevctl-cli/): jaidev posted jevctl, a friend's npm CLI that wraps TypeSafe's Jev as shell commands. verify, screen, classify, extract, find, rerank, match, route, ask, compact, and batch. Keys can be TypeSafe, OpenRouter, or Cloudflare. compact drops stale tool results the way Tran's sieve does, at a 0.5 keep threshold. Updated 2026-09-30. - [TypeSafe drops the Jev waitlist; new accounts start with $5](https://jevainews.com/news/jev-no-waitlist/): On October 8, 2026 the same signed-in console still showed available credits of $0.00. On September 20 TypeSafe dropped the Jev waitlist and posted $5 of credit. New signups no longer get it. On October 6, 2026 a new account showed $0.00. Updated 2026-10-08. - [TypeSafe posted an API and console outage, then said it was back](https://jevainews.com/news/typesafe-api-incident/): At 08:30 UTC on September 21, TypeSafe posted that the API and console had issues and pointed at status.typesafe.ai. At 09:53 UTC it posted "We are back!" The status incident is titled "Console is unavailable" and marked resolved at 08:16 UTC, before those tweets. Updated 2026-09-30. ## Scores and failures - [Evals](https://jevainews.com/evals/): Measurements already in the stories, each with the caveat its author published. - [Compare](https://jevainews.com/compare/): Similar tests placed next to each other, with each author's caveat. - [Limits](https://jevainews.com/limits/): Ways jev-1.13 fails, from TypeSafe's jaggedness list and from later stories. - [Decision model hallucination](https://jevainews.com/hallucination/): TypeSafe's September 15, 2026 launch post says Jev gives up string generation and can't hallucinate. Schema matching is the guarantee. Published hallucination tables stay on their own rows. - [imajev-4b](https://jevainews.com/imajev-4b/): Mohit Garg's Apache-2.0 LoRA on Qwen3.5-4B. Weights are on Hugging Face. Opened October 8, 2026, JevBench v1.6.1 lists capability 61.9 at rank 31, and Image JevBench v0.3.0 lists capability 68.0 at rank 1. Opened October 7, the open-weights list had capability rank 17. v1.5.6 lists 70.4 at rank 10. Image JevBench v0.1.5 lists 76.4 at rank 2. - [JevBench v1.5 ranks Cygnet 73.7 and Jev 1.13.0 72.1](https://jevainews.com/news/jevbench/): Opened October 19, 2026, JevBench JSON revision v1.6.29 ranks 184 open systems and 38 hosted ones. basal-omni-1.0-ultra is first of 184 on Capability at 82.5. kettei-fat stays first of 38 hosted at composite 77.8. Jev 1.14.0 is ninth at 70.2. Opened October 17, the API board named revision v1.7.46 and 39 hosted systems, with kettei-fat first at 77.8 and Jev 1.13.0 seventh at 71.5. Opened October 11, that board ranked 27 hosted systems, Sage 1.3.0 first at 74.0. JevBench v1.5.0, posted September 28, scores 1,624 decisions. The v1.5 page ranks Cygnet 73.7, Winnow-12B Q8 73.2, and Jev 1.13.0 72.1. Updated 2026-10-19. - [Open-Jev 27B v1.1 answers 197 of 231 public JevBench tasks](https://jevainews.com/news/open-jev-public/): Zefan Cai released Open-Jev 27B v1.1, a LoRA and decision head on Qwen3.8-27B. On the 231 public JevBench tasks it is correct on 197, three behind Jev 1.13.0 at 200. The hard tier is 80 of 111 against Jev at 81. The September 21 audit of the released 2B and 9B checkpoints, 150 and 179, is unchanged. The card prints no speed comparison. On October 2 one 64-step pilot on the released 2B checkpoint moved a 128-row synthetic test from 84 of 128 to 99 of 128. Those rows are not JevBench. Updated 2026-10-02. - [Deel's charts show 6 of 8 quality checks at parity, and two losses](https://jevainews.com/news/deel-production/): TypeSafe posted four charts on September 25 and attributed them to Deel. Six of eight quality checks are at parity or better. Messy metric questions fall 16.6 points. Three-level ticket tags fall 3.6. The live-shadow speed bar is 4 times. The tweet's 50% to 86% and 70% to 97% are not printed on the chart. We did not rerun it. Updated 2026-09-30. - [A registered attack on jev-1.13.0 moves validated success from 1.8% to 3.5%](https://jevainews.com/news/decision-hijack/): Tiantong Wu and Wei Yang Bryan Lim, NTU, posted arXiv:2609.28613 on September 23. On 510 reconstructed InjecAgent cases, an adaptive attack with score feedback lifts validated attack success on jev-1.13.0 from 1.8% to 3.5%. The mean highest attacker-target probability found during search goes from 0.043 to 0.084. Several registered phrasings do not help the attacker. The paper has no code repository in the text we read. We did not rerun it. Updated 2026-09-30. - [HiddenLayer prints Jev at 91% Confirm Price on a checkout ad, and this desk's jev-1.13.0 still chose place_order at 0.89](https://jevainews.com/news/hiddenlayer-jev/): HiddenLayer's October 14, 2026 write-up prints TypeSafe Jev at 91% Confirm Price and 9% Place order on a checkout ad. Date headers 07:24:57 GMT and 07:24:58 GMT October 19, this desk's jev-1.13.0 still chose place_order, 1.0 then 0.89. Updated 2026-10-19. - [OpenAI Decisions API is generally available; this desk's October 23 POSTs returned HTTP 200](https://jevainews.com/news/openai-jev/): On October 23, 2026, Date headers 02:23:12 GMT and 02:23:13 GMT, two POSTs from this desk to the Decisions API returned HTTP 200 from gpt-6-luna. OpenAI Developers posted general availability on October 22, 19:07 UTC. The guide prices uncached input at $0.10 per million tokens and cached input at $0.01, with no output charge. Requests now hold up to 400 questions. On October 8, 2026 ten urgency calls had a median clock of 668 ms on Jev and 387 ms on Decisions. Updated 2026-10-23. ## Newest stories - [OR-decision-4B prints JevBench 193 of 231, and Choice 79.7 against Jev 79.6 on alphaXiv's chart](https://jevainews.com/news/or-decision-4b/): alphaXiv posted OR-decision-4B on October 30, 2026, at 16:06 UTC. The Hugging Face card prints JevBench 193 of 231, 83.55%. The launch chart prints Choice 79.7 against Jev 79.6, with sarcasm at 53.6 against Jev 70.7. Apache-2.0 LoRA on Qwen3.5-4B. This desk did not run the weights. Updated 2026-10-31. - [BranchOne returns Choice probabilities from Qwen3.5-0.8B-Base, median 0.305 s on ten short text questions](https://jevainews.com/news/branchone/): Necati Demir posted BranchOne on October 29, 2026. The open checkpoint returns one probability per supplied candidate from Qwen3.5-0.8B-Base plus LoRA. His A100 Colab run prints median 0.305 s on ten short text questions. Apache-2.0. This desk did not run the weights. Updated 2026-10-31. - [MiniCPM5-2B-Jev prints 78.79% on JevBench, first among open models at or under 2B](https://jevainews.com/news/minicpm5-2b-jev/): OpenBMB posted MiniCPM5-2B-Jev on October 28, 2026, at 13:01 UTC. The project prints 78.79% on JevBench, first among open models at or under 2B, and 86.41% accuracy with ECE 0.0307 on decision-v7. Apache-2.0 on MiniCPM5-2B. This desk did not run the weights. Updated 2026-10-31. - [Opened October 26, Decision Index 0.4.1 Full score for d3-ultra is 70.18, second of 229](https://jevainews.com/news/d3-ultra/): Xunzhuo Liu posted d3-ultra on October 27, 2026, at 03:01 UTC. Opened October 26, Decision Index 0.4.1 Full score is 70.18, second of 229, +3.03 over d3 at 67.15. The card's private SR-Text mean is 87.2. Apache-2.0. This desk did not run the weights. Updated 2026-10-27. - [Decision Model Factory prints DF-JA-Large at 67.63 on S1MB Japanese Task Avg against a public snapshot high of 60.05](https://jevainews.com/news/decision-model-factory/): Yuma Ichikawa posted Decision Model Factory on October 27, 2026, at 03:19 UTC. arXiv 2610.34956 prints DF-JA-Large at 67.63 on S1MB Japanese Task Avg against a fixed public snapshot high of 60.05. No public weights on the pages this desk opened. This desk did not run a build. Updated 2026-10-27. - [hunch runs EmbeddingGemma2 as a CPU System One endpoint at 20 ms on text and 100 ms on images](https://jevainews.com/news/hunch/): Matt Mastracci's hunch turns EmbeddingGemma2 into a CPU System One endpoint at 20 ms on text and 100 ms on images. Cloudflare Clef teaches a local student in real time. Open-source launch was planned for later the same week. This desk has no public repo or hosted call yet. Updated 2026-10-28. - [Abe's 1,774 Snowflake posts cost $0.055 on Jev Noul against $1.87 on AI_FILTER](https://jevainews.com/news/snowflake-ai-filter/): ABE Masanori posted on October 24, 2026, at 11:20 UTC. On 1,774 of his 2025 X posts, Jev Noul cost about $0.055 against AI_FILTER at about $1.87, about 1/34. AUC against AI_FILTER was 0.9685. Of 70 splits where Jev scored under 0.2 and AI_FILTER was TRUE, he judged Jev right on 54. Date headers 08:49:12 GMT to 08:49:40 GMT October 26, this desk's jev-1.14.0 noul on his Lightdash example was 0.04. Updated 2026-10-26. - [Base Labs posts mammoth-1 at Decision Index 0.4 Full 71.24; live 0.4.1 prints 68.4](https://jevainews.com/news/mammoth-1/): Base Labs posted mammoth-1 on October 21, 2026, at 17:14 UTC, and the card prints Decision Index 0.4 Full 71.24, first of 208. Opened October 23, the live space is edition 0.4.1, 229 models, mammoth-1 Full 68.4 rank 3. Date headers 02:23:21 GMT to 02:23:23 GMT, GET /v1/models with no key returned HTTP 401. Updated 2026-10-31. - [Opened October 26, Decision Index 0.4.1 Full score for d3-turbo is 67.31, and vision full is 72.46](https://jevainews.com/news/d3-turbo/): Xunzhuo Liu posted d3-turbo on October 20, 2026, at 00:42 UTC. Opened October 26, Decision Index 0.4.1 Full score is 67.31 and vision full is 72.46, second of 56. The card's own 0.3.2 estimate is 70.2. Apache-2.0. This desk did not run the weights. Updated 2026-10-27. - [Sthānika's Sieve-31B card prints Decision Index 0.3.2 Full score 63.8](https://jevainews.com/news/sieve-31b/): Cuth posted Sieve 31B on October 20, 2026. The Hugging Face card prints Decision Index 0.3.2 Full score 63.8, board data of October 14, and a public index of 62.52 on the authors' own 0.3 run. Apache-2.0 LoRA on Gemma 4 31B. This desk did not run the weights. Updated 2026-10-20. - [Cloudflare posts Clairic, a System One extraction model at $0.10 per million input](https://jevainews.com/news/clairic/): Cloudflare posted Clairic on October 19, 2026. The blog's cascade ran in 19.3 s against 61.7 s for GLM-5.3 at every step, $6,032 against $11,740 per million claims. Docs price input at $0.10 per million tokens. This desk did not call the API. Updated 2026-10-20. - [Datadog prints Jev Noul recall 82.90% to 3.88% under a counterfeit policy, and this desk's is_risky fell from 0.94 to 0.13](https://jevainews.com/news/datadog-jev/): Datadog Security Labs' October 15 write-up, posted October 19, prints Jev Noul recall 82.90% on raw attacks and 3.88% with a counterfeit policy prepended. Date headers 08:50:46 GMT and 08:50:47 GMT October 20, this desk's jev-1.13.0 is_risky was 0.94 then 0.13. Updated 2026-10-20. ## Site index - [Home](https://jevainews.com/): Featured story plus the twelve newest others. - [News](https://jevainews.com/news/): Every story, newest article date first. - [Timeline](https://jevainews.com/timeline/): The same stories grouped by article date. The date is the primary post or write-up. - [Topics](https://jevainews.com/topics/): Stories grouped by tag. - [People](https://jevainews.com/people/): People whose posts this digest rewrote. - [Repos](https://jevainews.com/repos/): GitHub projects cited in the stories, including community stand-ins. - [Terms](https://jevainews.com/terms/): Definitions written in the notes on each article. - [Privacy](https://jevainews.com/privacy/): Google Analytics on the public site, advertising cookies when ads are shown, and Google's AdSense consent message for the EEA, UK, and Switzerland. Contact is hi@jevainews.com. - [Sources](https://jevainews.com/sources/): Original URLs cited at the bottom of the stories. - [RSS](https://jevainews.com/rss.xml): Newest stories as a feed. - [Sitemap](https://jevainews.com/sitemap-index.xml): Every public URL on the site. - [Full text](https://jevainews.com/llms-full.txt): Body, sources, desk notes, and update date for every story. ## Topics - [Agents](https://jevainews.com/topics/agents/): Loops where Jev picks the next action and another system carries it out. 17 stories. - [Browser](https://jevainews.com/topics/browser/): Page loops where Jev picks an action and a browser runtime performs it. 5 stories. - [Demos](https://jevainews.com/topics/demos/): Single runs and demonstrations that are not a scored benchmark. 18 stories. - [Ecosystem](https://jevainews.com/topics/ecosystem/): Gateways, plugins, and hosted doors around the TypeSafe API. 39 stories. - [Evals](https://jevainews.com/topics/evals/): Measurements that print a sample size, a table, or the author's caveat. 117 stories. - [Explainers](https://jevainews.com/topics/explainers/): What a System One model returns, and the words this desk uses for those returns. 7 stories. - [Launch](https://jevainews.com/topics/launch/): TypeSafe's own announcements, from the September 15 launch through later access changes. 10 stories. - [local](https://jevainews.com/topics/local/): 2 stories tagged local. - [measurement](https://jevainews.com/topics/measurement/): 5 stories tagged measurement. - [Open source](https://jevainews.com/topics/open-source/): Open-weight stand-ins and local ports, with the numbers their authors published. 84 stories. - [Product](https://jevainews.com/topics/product/): Products that call Jev, including routers, CLIs, warehouses, and SQL functions. 127 stories. - [research](https://jevainews.com/topics/research/): 5 stories tagged research. ## Optional - [Nace posts Drex Max at Decision Index 0.3.2 public 67.59, $0.06 per million input](https://jevainews.com/news/drex-max/): Nace posted Drex Max on October 19, 2026. Updated 2026-10-23. - [Qwen-Audio-Jev prints 78.70% on MMAU-mini and about 20x the base model's speed](https://jevainews.com/news/qwen-audio-jev/): Qwen-Audio-Jev prints 78.70% on MMAU-mini and a 20x speedup against Qwen-Audio-3.1-ASR-Flash-Next. Updated 2026-10-19. - [ToddlerJev prints 68.70% on MMAU-mini, with weights under the Qwen Research License](https://jevainews.com/news/toddlerjev/): ToddlerJev, posted October 18, 2026, prints 68.70% on MMAU-mini, 61.30% on MMSU, and 56.90% on MMAR. Updated 2026-10-19. - [Opened October 19, JevBench ranks basal-omni-1.0-ultra first of 184 open systems on Capability at 82.5](https://jevainews.com/news/basal-omni/): Opened October 19, 2026, JevBench ranks basal-omni-1.0-ultra first of 184 open systems on Capability at 82.5. Updated 2026-10-19. - [Opened October 17, JevBench API board v1.7.46 ranks kettei-fat first of 39 at composite 77.8](https://jevainews.com/news/kettei-fat/): Opened October 17, 2026, JevBench API board v1.7.46 ranks kettei-fat first of 39 at composite 77.8. Updated 2026-10-17. - [Opened October 16, Artificial Analysis Decision Index prints d1-ultra at 65 and Jev 1.14 at 57](https://jevainews.com/news/d1-ultra/): Vals AI, October 22, 2026, 23:57 UTC, printed d1-ultra as the only decision model to get all 1,960 ClaimBench claims right. Updated 2026-10-23. - [On 492 tool calls, OpenRouter Armor's default Jev 1.13 scored 90.7%, 446 of 492](https://jevainews.com/news/openrouter-armor/): On 492 tool calls, OpenRouter Armor's default Jev 1.13 scored 90.7%, 446 of 492, at $0.0021 a call. Updated 2026-10-16. - [Span-02's card prints Decision Index 0.3 text index 63.62, and the blog prints thinking off at 61.2 against Jev 58.0](https://jevainews.com/news/span-02/): Hendrix Liu posted October 16, 2026, 23:16 UTC, that Span-02 had more usage in under 24 hours than Span-01 did in 20 days. Updated 2026-10-17. - [Jeva-4B prints 0 flips on 600 Choice permutations, and BANKING77 at K=16 is 91.62 against Jev 82.50](https://jevainews.com/news/jeva/): Hugging Face DailyPapers posted Jeva on October 14, 2026, at 20:10 UTC. Updated 2026-10-26. - [On DeepEval's Paris weather example, jev-1.13.0 scored 0.254 when the answer invented a breeze](https://jevainews.com/news/jeveval/): On October 14, 2026, at 01:58 UTC, two POSTs to jev-1.13.0 scored DeepEval's invented-breeze case at 0.254. Updated 2026-10-14. - [On 40 client requests, Jev and Opus both scored 37 of 40, at $0.0015 against $0.005 to $0.015](https://jevainews.com/news/trueforge-jev/): Akshay Pachaar's TrueForge agent scored 37 of 40 client-issue matches with Jev and 37 of 40 without it. Updated 2026-10-14. - [On eight Stripe passages, a 0-9 Score put Roll secret at 8.86 and the other six at 2.42 or below](https://jevainews.com/news/choice-bradley-terry/): On eight Stripe passages this desk sent to jev-1.13.0, a 0-9 Score put the Roll secret steps at 8.86 and the other six at 2.42 or below. Updated 2026-10-12. - [SubQ Snap 1.1 prints 65 on Artificial Analysis Decision Index, $0.04 per million input](https://jevainews.com/news/subq-snap/): On October 23, 2026, Date headers 02:23:17 GMT to 02:23:20 GMT, three POSTs from this desk to SubQ Snap returned HTTP 200. Updated 2026-10-23. - [Opened October 26, Decision Index 0.4.1 Full score for d3 is 67.15; d3-turbo and d3-ultra have their own pages](https://jevainews.com/news/decision-3-0/): Opened October 26, 2026, Decision Index 0.4.1 Full score for d3 is 67.15. Updated 2026-10-27. - [Mizan 4B averages 78.7 on 3,000 Persian records](https://jevainews.com/news/mizan-4b/): xmanii/Mizan-4B prints a mean accuracy of 78.7 on 3,000 real Persian records, against the card's Jev cell of 83.7. Updated 2026-10-12. - [THX-01 averages 98.4 on its 2,843 support tickets](https://jevainews.com/news/thx-01/): HAL-X's THX-01 card averages 98.4 on 2,843 support tickets, against Jev 1.13 at 97.4 on the same four sets. Updated 2026-10-12. - [PurpleMIST-Flash-1.0 scores 0.668 on 15,050 translated African-language decisions](https://jevainews.com/news/purplemist-flash/): PurpleMIST-Flash-1.0 scores 0.668 accuracy on 15,050 translated African-language decisions, against DeepSeek-V4-Pro at 0.669. Updated 2026-10-10. - [JevBench ranks Sage 1.3.0 first of 27 hosted systems at composite 74.0](https://jevainews.com/news/sage/): Opened October 11, 2026, the JevBench API board ranks Sage 1.3.0 first of 27 hosted systems at composite 74.0. Updated 2026-10-10. - [TypeLLM scores Qwen3.8-27B at 228 of 231 public JevBench tasks when thinking is on](https://jevainews.com/news/typellm/): On October 16, 2026, at 05:44 UTC, TypeLLM posted that Jev 1.14 cannot return an array of vehicle boxes, including Tesla's flying Roadster. Updated 2026-10-16. - [Vega, the vegaml package, scores 75.4% on 800 phishing emails against Jev at 61.9%](https://jevainews.com/news/vegaml/): Nandha Kishor M's Vega, the vegaml package, scores 75.4% on 800 phishing emails against Jev 1.13.0 at 61.9%. Updated 2026-10-10. - [On WANDS, more Jev questions move the mean to 0.610 and leave the median at 0.561](https://jevainews.com/news/bag-of-decisions/): Doug Turnbull's WANDS table prints mean NDCG@10 of 0.610 for a bag of Jev questions and 0.574 for one question. Updated 2026-10-10. - [Clef-omni scores a 21-second video in about 1.5 seconds, at $0.15 per million input tokens](https://jevainews.com/news/clef-omni/): Cloudflare's October 9, 2026 blog prices Clef-omni at $0.15 per million input tokens. Updated 2026-10-10. - [Microsoft-Decision-1 lists $0.042 per million input tokens and calls itself 4.5 times quicker than Quyet-1.0-Large](https://jevainews.com/news/microsoft-decision-1/): Opened October 11, 2026, the JevBench API board ranks Microsoft-Decision-1 sixth of 27 hosted systems at composite 69.1. Updated 2026-10-11. - [OpenBles Firewall 1 catches 119 of 120 agent attacks and flags no benign look-alike](https://jevainews.com/news/openbles-firewall/): OpenBles Firewall 1 caught 119 of 120 agent attacks and flagged none of 120 benign look-alikes on its held-out test. Updated 2026-10-10. - [Celeris-1 Decision prints 80.0% on 3,210 jev-bench items at a 67 ms median, and Artificial Analysis lists it at 52](https://jevainews.com/news/celeris-1-decision/): Celeris-1 Decision, posted October 8, 2026, prints 80.0% on 3,210 jev-bench items at a 67 ms median, and $0.04 per million input tokens. Updated 2026-10-16. - [Open decision models start from an open language model](https://jevainews.com/news/open-decision-bases/): Open decision checkpoints start from a public language model because a closed API does not release the weights a decision head has to read. Updated 2026-10-08. - [Step-Jev scores 19.6% when the reward grades every search step](https://jevainews.com/news/step-jev/): On October 9, 2026 hosted jev-1.13.0 scored the repository's no-step test state incorrect at probability 1.0. Updated 2026-10-09. - [H2O-Lightning-4B v1.1 prints JevBench composite 72.5 against Jev at 71.5](https://jevainews.com/news/h2o-lightning-4b/): H2O's October 7, 2026 blog prints H2O-Lightning-4B v1.1 at JevBench composite 72.5, against Jev 1.13.0 at 71.5. Updated 2026-10-10. - [CLERC top-1 moves from 5% to 18% when jev-1.12 reranks the shortlist](https://jevainews.com/news/jev-rag/): TypeSafe's CLERC cookbook moves top-1 from 5% to 18% when jev-1.12 reranks a BM25 shortlist of 30 on 40 queries. Updated 2026-10-08. - [Arena scores typesafe/jev-router at +3.6% and $0.155 a task](https://jevainews.com/news/jev-router/): Arena's October 7, 2026 thread scores typesafe/jev-router at +3.6% net and $0.155 a task. Updated 2026-10-08. - [Open d1's d1-3B card prints 48.57 on a self-run of Decision Index 0.2.1](https://jevainews.com/news/open-d1/): On October 7, 2026 Liquid AI posted Open d1, and the d1-3B card prints 48.57 on a self-run of Decision Index 0.2.1. Updated 2026-10-08. - [PolicyLM-1.7B scores each policy category from 0 to 1](https://jevainews.com/news/policylm-1-7b/): PolicyLM-1.7B prints custom-policy accuracy 0.842 at a 0.5 cutoff, and it returns a score from 0 to 1 for each category, up to 16 in one pass. Updated 2026-10-07. - [Vela 2.0's 9B card prints 41.63 on Decision Index 0.2.1](https://jevainews.com/news/vela-2-0/): The Vela 2.0 9B card prints Decision Index 0.2.1 at 41.63 with Noul calibration on, and 41.09 with it off. Updated 2026-10-08. - [Blink v0.3 prints Decision Index 0.2.1 at 57.48](https://jevainews.com/news/blink/): Pixilab's Blink v0.3 card prints a self-run of Decision Index 0.2.1 at 57.48, against the board's Jev cell of 57.91. Updated 2026-10-12. - [Nace posts Drex 1.1, and the card titles the weights Drex DLM](https://jevainews.com/news/drex-dlm/): Nace posted Drex 1.1 on October 6, 2026, and the card scores Decision Index 0.2 at 52.31 against the kit's Jev cell of 51.67. Updated 2026-10-07. - [fsdecide scores 91.3% on ForkScape review, and Jev scores 56.3%](https://jevainews.com/news/fsdecide/): ForkScape posted fsdecide on October 5, 2026, a 149M ModernBERT trained on the app's own decisions. Updated 2026-10-05. - [Quyet-1.0-Large's homepage prints JevBench v1.6.0 at 71.39](https://jevainews.com/news/quyet/): Quyet-1.0-Large's homepage prints JevBench v1.6.0 official score 71.39, rank 1 of 92, and Capability 81.7. Updated 2026-10-12. - [basal-1.5 scores 72.12% on Werdykt, and Jev 1.13.0 scores 81.64%](https://jevainews.com/news/basal-1-5/): Remek Kinas posted basal-1.5 on October 4, 2026. Updated 2026-10-19. - [JEV-27B-VL finishes 15 of 20 pick-and-place scenes](https://jevainews.com/news/jev-27b-vl/): AutoTrust posted a JEV-27B-VL follow-up on October 4, 2026. Updated 2026-10-05. - [Decision 2.0's Vega card prints 56.5 on its own Decision Index run](https://jevainews.com/news/decision-2-0/): vLLM Semantic Router posted Decision 2.0 on October 3, 2026. Updated 2026-10-11. - [AutoTrust scores GEV-26B-Decide at 62.48 on its own Decision Index run](https://jevainews.com/news/gev-26b-decide/): AutoTrust posted GEV-26B-Decide on October 3, 2026, a Gemma 4 26B model with about 4B active. Updated 2026-10-03. - [NIRNAY 450M prints 0.8792 on 3,080 Banking77 rows and says the Jev cell was not measured here](https://jevainews.com/news/nirnay-450m/): Gautam Kishore posted NIRNAY 450m on October 3, 2026, an Apache-2.0 Laya fine-tune from Eulogik. Updated 2026-10-03. - [StartLux-Decision v1.1 27B Instant prints Full score 69.12, first of 174 on Decision Index 0.3.2](https://jevainews.com/news/startlux-decision/): The StartLux-Decision README, opened October 20, 2026, prints v1.1 27B Instant at Full score 69.12, first of 174 on Decision Index 0.3.2. Updated 2026-10-20. - [WaterSheep fine-tunes ModernBERT and serves /v1/systemone with a multi-label question](https://jevainews.com/news/watersheep/): Samrat Dutta posted WaterSheep on October 3, 2026. Updated 2026-10-03. - [Bud Decision Studio serves eleven local decision models on port 8420](https://jevainews.com/news/bud-decision-studio/): Bud Decision Studio runs eleven open decision models on a local TypeSafe-shaped API at port 8420, and its README prints a GB10 training table. Updated 2026-10-02. - [A CUA Speedrun chart prints Jev at 9.8% on OSWorld-50, 125 seconds a task](https://jevainews.com/news/cua-speedrun/): Pranjal Aggarwal posted a CUA Speedrun chart on October 2, 2026. Updated 2026-10-03. - [PageIndex walks long filings with Jev Choice calls and checks two answers](https://jevainews.com/news/jev-doc-search/): VectifyAI's jev-doc-search, Apache 2.0, asks jev-1.13.0 to walk a PageIndex tree, and on two annual reports the kept pages include the income statement. Updated 2026-10-02. - [A gradient-boosting ensemble beats Jev 1.13 on five CYP assays](https://jevainews.com/news/mmai-cyp/): On 12,563 molecules, an Insilico Medicine chart puts a gradient-boosting ensemble above Jev 1.13 on five CYP assays, and Jev's probabilities stop at 0.69. Updated 2026-10-02. - [StreamDecisionBench puts Kev-27B at 62.15 and Jev at 59.63](https://jevainews.com/news/stream-decision-bench/): Jacob Lin posted StreamDecisionBench on October 2, 2026. Updated 2026-10-03. - [Cloudflare posts Clef and Clef-flash, Apache 2.0 decision models on Workers AI](https://jevainews.com/news/cloudflare-clef/): On October 9, 2026 Cloudflare's blog prices Clef-flash at $0.038 per million input tokens, down from the $0.09 on the October 1 card. Updated 2026-10-28. - [Databricks posts ai_decide, a Beta function that returns a noul, a choice, or a score](https://jevainews.com/news/databricks-ai-decide/): Databricks posted on October 16, 2026 that TypeSafe's Jev-1.14 is available through ai_classify and ai_decide on the Lakehouse. Updated 2026-10-17. - [DSPy 3.4.0 calls Jev from Predict, with a gate at 0.2 and 0.6](https://jevainews.com/news/dspy-jev/): DSPy 3.4.0 adds experimental Noul, Choice, and Score types plus a TypeSafe client, and a patterns repo pins jev-1.13.0. Updated 2026-10-02. - [GLiDE v1.2 prints Decision Index 0.4.1 Full 71.5, first of 229, at $0.04](https://jevainews.com/news/glide/): Fastino posted GLiDE v1.2 on October 22, 2026, at 16:36 UTC. Updated 2026-10-26. - [HydroJEV tests jev-1.13.0 as a one-second screen on simulated water-network alarms](https://jevainews.com/news/hydrojev/): Tianwei Mu and coauthors test Jev as a first screen on simulated water-network alarms, and a gate settles 69 of 200 and 76 of 200 windows. Updated 2026-10-02. - [A frozen Jev pipeline reports the lowest ten-corpus valence-arousal error in a SemEval table](https://jevainews.com/news/jev-dimabsa/): Yiqun Zhang and coauthors report the lowest ten-corpus valence-arousal error among full-coverage systems, using frozen Jev and 488 fitted coefficients. Updated 2026-10-02. - [Perplexity posted pplx-decider-v1.2 on October 16, with Decision Index 0.3 overall 68.09 on joint requests against v1.1 at 62.30](https://jevainews.com/news/pplx-decider/): Perplexity posted pplx-decider-v1.2 on October 16, 2026, with Decision Index 0.3 overall 68.09 on joint requests against v1.1 at 62.30, at $0.02 per million input tokens. Updated 2026-10-17. - [You.com's risk sweeps cost $0.14 each, and the report means tie](https://jevainews.com/news/risk-analysis-server/): Edward Irby's three escalated sweeps cost $0.43 in all, and a blind read of nine Jev reports and six Qwen reports ties at a mean of 4.25. Updated 2026-10-02. - [Strands Decider 2B scores 167 of 231 public JevBench tasks](https://jevainews.com/news/strands-decider/): Strands Labs posted Decider 2B on October 1, 2026, scoring 167 of 231 public JevBench tasks at a median 115 ms on an RTX 3090. Updated 2026-10-02. - [Nace posts Drex 1.5 at $0.04, and its chart shows 199 of 231 against Jev's 201](https://jevainews.com/news/drex-1-5/): The October 9, 2026 weights card prints Drex v1.5 at Decision Index 0.3.1 public index 58.08, against Jev 1.13.0 at 57.96. Updated 2026-10-11. - [The Jeeves README prints 0.865 on the public JevBench hard tier](https://jevainews.com/news/jeeves/): PostHog/jeeves, MIT, is a Qwen3.5-9B LoRA plus a pointer head. Updated 2026-09-30. - [TechCrunch prices one Jev monitor at $2.94, and the demo page prints another estimate](https://jevainews.com/news/jev-sentinel/): Tim Fernholz, TechCrunch, September 30, describes Shapor Naghibzadeh's Jev Sentinel demo and writes that this kind of monitoring costs $2.94 with Jev against $372 with a frontier LLM, in theory. Updated 2026-10-01. - [jgrep's test bench prints mean recall 0.926 across 60 commits, at $0.1112](https://jevainews.com/news/jgrep-tests/): kyu1204/jgrep, MIT, npm name jevgrep, command jgrep, is a different project from dzhng/jevgrep. Updated 2026-09-30. - [Maincode's Matilda card prints 59.26 on a self-run of Decision Index 0.2.1](https://jevainews.com/news/matilda-jev/): Opened October 8, 2026, the Matilda card prints Decision Index 0.3 at 60.16, and the model API lists private false. Updated 2026-10-08. - [Inception calls Mercury Decide first on OpenRouter's JevBench v1.4, and prints no score](https://jevainews.com/news/mercury-decide/): Opened October 8, 2026, the JevBench API board ranks Mercury Decide third at composite 72.4. Updated 2026-10-08. - [Ollaya serves local decision models on port 11435, and prints winnow:e4b at 0.722](https://jevainews.com/news/ollaya/): ollaya-dev/ollaya, Apache-2.0, calls itself Ollama for decision models and says it is independent of Ollama and of TypeSafe. Updated 2026-09-30. - [Wrangle's chart prints 137.3 perfect matches per 200 with Jev](https://jevainews.com/news/wrangle-bench/): John Crown's WrangleBench chart prints 137.3 perfect matches per 200 with Jev and 105.1 without, and the post calls that a 30.64 percent gain. Updated 2026-10-02. - [DiffusionGemma as Jev reads one denoising step, and a hand chart sums to 198 of 201](https://jevainews.com/news/diffusiongemma-as-jev/): Google's Gemma account pointed at vLLM pull 57250 on September 29: seed a canvas, read yes/no, choice, and score probabilities in one denoising step. Updated 2026-10-07. - [A 37-dataset paper scores frozen jev-1.13.0 against two open models](https://jevainews.com/news/jev-benchmarking/): Tobias Deußer, Lorenz Sparrenberg, and Rafet Sifa submitted arXiv:2609.37647 on September 29, 2026. Updated 2026-10-03. - [A v3 judge paper puts a frozen Jev cascade 0.9 points above GPT-6 on 1,610 pairs](https://jevainews.com/news/jev-judge-cascade/): Yubo Li, Yidi Miao, Ramayya Krishnan, and Rema Padman revised arXiv:2609.26550 to v3 on September 29. Updated 2026-09-30. - [JevScout ranks oversized MCP text and prints 38% less agent time on its own check](https://jevainews.com/news/jevscout/): Tim Lin posted JevScout on September 29. Updated 2026-09-30. - [Liquid's chart puts d1 at 58.9 on an internal Decision Index 0.2.1](https://jevainews.com/news/liquid-d1/): Liquid AI posted a chart on September 29 captioned as an internal reproduction of Decision Index 0.2.1. Updated 2026-10-16. - [OpenAI's Decisions API preview says results come back in 150 milliseconds](https://jevainews.com/news/openai-decisions/): On October 22, 2026 OpenAI Developers posted the Decisions API generally available, and this desk's October 23 POSTs returned HTTP 200. Updated 2026-10-23. - [A Pi extension asks Jev which skills to load, and the 20-skill demo drops about 1,000 tokens](https://jevainews.com/news/pi-skill-picker/): Arnav Gupta posted a Pi skill picker on September 29. Updated 2026-09-30. - [Upstage's Solar Decide chart prints 87.0% on the public 231, and a p95 of 15.7 seconds](https://jevainews.com/news/solar-decide/): Sung Kim posted Solar Decide on OpenRouter on September 29. Updated 2026-09-30. - [Violoop ran the same desktop harness with Jev and with Opus 5.5 on one QuickBooks bill](https://jevainews.com/news/violoop-live/): Violoop posted four live desktop runs on September 29. Updated 2026-09-30. - [Glean replayed four internal decisions on Jev, and 40 expert transfers were 8.1 times faster](https://jevainews.com/news/glean-four/): Tony Gentilcore posted Glean's write-up on September 28. Updated 2026-09-30. - [One day of grant posts, 100 of them, hid 75 of 78 rejects after the questions changed](https://jevainews.com/news/grants-100/): Deji labeled 100 funding posts himself, then scored Jev against that key. Updated 2026-09-30. - [imajev-4b: Hugging Face weights and JevBench 70.4](https://jevainews.com/news/imajev/): Opened October 8, 2026, the JevBench v1.6.1 open-weights list puts imajev-4b at capability rank 31 and 61.9. Updated 2026-10-08. - [JevAdvBench's abstract says one appended opinion flips 12.1% of jev-1.13.0 decisions](https://jevainews.com/news/jevadvbench/): Jianyi Hu and coauthors submitted arXiv:2609.31142 on September 25. Updated 2026-09-30. - [jevbench.app replays recorded Jev answers on labeled examples, and one test is 20 of 21](https://jevainews.com/news/jevbench-app/): Yusuke posted jevbench.app, a playground of unit tests for questions you have already labeled. Updated 2026-10-03. - [jevgrep solves the same 8 of 10 SWE-bench tasks, and the bill including Jev is 25.8% lower](https://jevainews.com/news/jevgrep/): David posted dzhng/jevgrep, MIT, a CLI that asks Jev which files matter and returns excerpts. Updated 2026-09-30. - [A 400M encoder named Kev scores 79.96% against Jev on 17,476 of its own rows](https://jevainews.com/news/kev-400m/): Sandipan Kundu posted skundu42/kev, a 400M encoder with Choice, Noul, and Score. Updated 2026-09-30. - [Polylane moved five on-call decisions to Jev and prints a 39.1% lower bill per 1,000 calls](https://jevainews.com/news/polylane-oncall/): Vi Tran posted Polylane's write-up on September 28. Updated 2026-09-30. - [Rox's chart keeps 96% of relevance at 200 chunks, against 84% in production](https://jevainews.com/news/rox-rerank/): Rox posted on September 28 a chart titled Jev vs. our production reranker, a mean over 75 queries. Updated 2026-10-01. - [toolgate 0.15 separates writing a script from running it, 19 of 20 on the paired set](https://jevainews.com/news/toolgate/): Risk Averse Technology Company posted toolgate 0.15, MIT, a Claude Code hook and MCP proxy that asks Jev before a risky tool call. Updated 2026-09-30. - [A DTap chart puts a Jev 1.13 tool selector at 70.1% attack success, and a second Jev call at 36.1%](https://jevainews.com/news/dtap-jev/): Zhaorun Chen posted a DTap chart on September 27 for Jev 1.13 used as a tool selector. Updated 2026-09-30. - [Plumb-4B reaches 89 of 111 public hard items, and the official board scored it 65.84](https://jevainews.com/news/plumb-4b/): Christopher House published crh225/plumb, Apache-2.0, starting from JevK5 v0.2. Updated 2026-09-30. - [astra-jev-harness cuts Astra input 27% on one two-file fixture](https://jevainews.com/news/astra-jev/): Louis posted Oranquelui/astra-jev-harness, MIT, version 0.5.0. Updated 2026-09-30. - [jev-code's three-task ladder still fails after the call-nesting fixes](https://jevainews.com/news/jev-code/): Roberto Montalti posted rhighs/jev-code on September 26. Updated 2026-09-30. - [jevos-1b prints 0.815 on 2,000 yes/no questions, against Jev at 0.927](https://jevainews.com/news/jevos/): Federico Elia released feder-cr/jev, a local server called jevos-1b, built with Loris Salsi. Updated 2026-09-30. - [RLCDAlignBench puts a generic Noul at median AUROC 0.886, and median ECE at 0.168](https://jevainews.com/news/rlcd-align/): Ruoqi Guo, Yi Liu, and seven coauthors posted RLCDAlignBench on September 24. Updated 2026-09-30. - [Decision Index 0.2 prints Jev at 51.67, and a later post names another first place](https://jevainews.com/news/decision-index/): Opened October 8, 2026, the Decision Index space still headers v0.3, 115 models, updated 2026-10-07, and Perplexity Decider v1.1 is still first at 62.8. Updated 2026-10-12. - [Jevrail scores a shell command before it runs, and the printed corpus leaves the model off](https://jevainews.com/news/jevrail/): Saman Pandey posted jevrail, an MIT hook that asks Jev whether a shell command is likely to destroy something, then blocks, asks, or allows it. Updated 2026-09-30. - [Lightpanda's hostel script counts browser memory, and the Jev clock stays in the post](https://jevainews.com/news/lightpanda-hostel/): Lightpanda posted that 129 of its browser agents fit on one laptop against 12 Chrome agents, with 11 times less memory on a booking site, and that a hostel task took 6.7 seconds with Jev against 14.1 seconds with gemini-3.8-flash. Updated 2026-09-30. - [Eikos-27B prints 82.9 on its JevBench hard row, and 2.4% error once confidence is 0.90](https://jevainews.com/news/eikos/): Caio Vicentino released Eikos-4B and Eikos-27B under MIT. Updated 2026-09-30. - [mizchi's fit notes record 0 illegal moves in 489 MOBA calls, and 6 to 27 wrong pages where a one-line rule was 16/16](https://jevainews.com/news/mizchi-fit/): mizchi posted a fit table for Jev on September 24. Updated 2026-09-30. - [On Ámbar Reflex the 4B logit player went further than the more accurate 9B](https://jevainews.com/news/ambar-reflex/): Abel Coronado's local server plays an original run-and-gun with six typed questions and no generated text. Updated 2026-09-30. - [Five runs leave a 3.3-point lead on 18 claim checks](https://jevainews.com/news/claim-support-42/): Jamie Watters reran a 42-claim support check five times on September 22. Updated 2026-09-30. - [CLM-8B caches the action list, and a fine-tuned head scores 31 of 38 DeepSWE holdouts](https://jevainews.com/news/clm-8b/): Jacky Kwok posted CLM-8B on September 23. Updated 2026-09-26. - [Jev keeps the same direction on 28,074 classical-Korean sentences](https://jevainews.com/news/dansa-judges/): vase.park pointed at a Jev rerun of 28,074 translated sentences in hw725/dansa-research. Updated 2026-09-30. - [JEV Search moves NFCorpus nDCG@10 from 0.322 to 0.377](https://jevainews.com/news/jev-file-search/): andre posted assadiandre/jev-search, a macOS app that ranks files with no saved index. Updated 2026-09-30. - [jev_local's Sarashina row beats LFM on two JGLUE tests, and the README says that is not a ranking](https://jevainews.com/news/jev-local-sarashina/): あるごす added Sarashina2.2 Vision 3B to a local /v1/systemone server. Updated 2026-09-30. - [jevmod scores sexual text at 0.984 AUROC, and ZeroNSFW keeps the pictures local](https://jevainews.com/news/jev-nsfw/): jevmod's sexual AUROC on OpenAI's moderation eval is 0.984. Updated 2026-10-03. - [A 100,000-line needle takes 18.0 minutes with Jev and 24.7 without](https://jevainews.com/news/jev-rlm-needle/): El Asombroso Nosequien posted elalber2000/jev-rlm, a fork of a minimal recursive-language-model loop. Updated 2026-09-30. - [JevGate turns one function at a time into a finding at 0.80](https://jevainews.com/news/jevgate/): Nodehaven Online posted jevgate, a Rust review gate on crates.io as v0.5.1. Updated 2026-09-30. - [Jevonian asks Jev which model should take the next coding turn](https://jevainews.com/news/jevonian/): Chen posted xinyao27/jevonian, an AGPL-3.0 local proxy on port 8787. Updated 2026-09-30. - [Jevre's test split is 0.9323, and held-out sources fall to 0.8366](https://jevainews.com/news/jevre/): Di Zhang released a 0.85B checkpoint that speaks the TypeSafe Python SDK from localhost. Updated 2026-10-07. - [Jevsus ran a true/false diode on jev-1.13.0, and 45% of statements move with the wording](https://jevainews.com/news/jevsus/): 2nd1st published every raw answer from a September 20 run of jev-1.13.0. Updated 2026-09-30. - [One post scores Jev at 100 and Laya's English checkpoint at 92, on 100 items](https://jevainews.com/news/laya-hundred/): 码农暖爸 posted a 100-item English judgment set. Updated 2026-10-07. - [One Jev pass scores 1,072 translations for $0.3165](https://jevainews.com/news/multilingual-scores/): 七誌 rebuilt a translation bake-off so Jev 1.13.0 scores each language once. Updated 2026-09-30. - [One screenshot decision was $0.0002 on Jev and $0.032 on Opus 5](https://jevainews.com/news/ocr-computer-use/): awlevin's macOS loop reads the screen with OCR and asks Jev which action comes next. Updated 2026-09-30. - [postgres-search can sink a bad hit, and the README has not measured that step](https://jevainews.com/news/postgres-search/): Alex posted an Algolia-style search in Postgres with an optional Jev pass. Updated 2026-09-30. - [Rizzo Flow is 49 ms on an RTX 5060 Ti, and six missing-evidence rows stay confident](https://jevainews.com/news/rizzo-flow/): Simone Rizzo's Apache-2.0 server reads Spark-X2.5 logits through llama.cpp and speaks POST /v1/systemone. Updated 2026-09-30. - [Together prices Tev1 input at $0.04, and Ollama serves Choice, Score, and Noul](https://jevainews.com/news/tev1/): Opened October 7, 2026, Together's Tev1 page lists input at $0.04 per million tokens and free output. Updated 2026-10-07. - [AgentOS posts a 121-turn routing table for an opt-in Jev strategy](https://jevainews.com/news/agentos-pilot-router/): AgentOS added Jev as a third Pilot Router strategy, off by default. Updated 2026-09-30. - [ASO Workbench ranks App Store terms with a model trained on Jev's answers](https://jevainews.com/news/aso-workbench/): Mustafa Kendigüzel posted ASO Workbench, a Mac app that estimates Apple search popularity. Updated 2026-09-30. - [A 612-row banking-feed test puts Jev under two Gemini rows on accuracy](https://jevainews.com/news/banking-feed-612/): Peter posted a categorization run on 612 sanitized rows from a private banking-feed set of about 33,000. Updated 2026-09-30. - [meraGPT posts Decider 1 ahead of Jev on a teacher-labelled set](https://jevainews.com/news/decider-1/): Latent Node posted Decider 1 (sd-1), a hosted model that answers noul, choice, and score questions in one call. Updated 2026-09-30. - [dejevu books the same Flights search faster than the published Jev demo, on another machine](https://jevainews.com/news/dejevu-flights/): Idov Mamane posted dejevu, an MIT browser loop that sends one JSON action to an ordinary chat model. Updated 2026-09-30. - [A Cloud Run recipe serves DiffusionGemma-Jev at 35 to 60 ms](https://jevainews.com/news/djev-run/): Google's Gemma account posted a one-command Cloud Run deploy for DiffusionGemma-Jev. Updated 2026-09-30. - [Exa posts a demo that puts web results into Jev's state](https://jevainews.com/news/exa-jev-web/): Ishan Goswami posted a demo at demos.exa.ai/jev-web-context. Updated 2026-09-30. - [hotdog-27B leads a 1,631-decision holdout and trails on 74 public items](https://jevainews.com/news/hotdog-27b/): Michael posted hotdog-27B, a rank-16 LoRA on Qwen3.8-27B that reads two logits and returns yes or no. Updated 2026-09-30. - [jev-gateway asks Jev which tool a coding agent should call](https://jevainews.com/news/jev-gateway/): Vini Lana posted jev-gateway, an MIT local proxy for Codex, Claude Code, OpenCode, and Gemini CLI. Updated 2026-09-30. - [jev-rubiks asks Jev when to speak, after a test where picking moves did nothing](https://jevainews.com/news/jev-rubiks/): maxlibin/jev-rubiks is an MIT browser cube. Updated 2026-09-30. - [jevelry turns a Jev call into a folder, a verdict, and a log](https://jevainews.com/news/jevelry/): Pavel Hegler posted backant-io/jevelry, an MIT npm runtime. Updated 2026-09-30. - [jym asks Jev which real subcommand a mistyped one was aiming at](https://jevainews.com/news/jevyoumean/): syumai posted jevyoumean, an MIT Go wrapper. Updated 2026-09-30. - [A Japanese table has Jev at 96.0% on one set and 56.0% on JamC-QA](https://jevainews.com/news/jp-benchmarks/): t.toda posted a three-column table. Updated 2026-09-30. - [jsort ranks a file by asking Jev which of two texts is more of something](https://jevainews.com/news/jsort-pairwise/): Khaled Eltokhy published jsort, an MIT command that fits a Bradley-Terry scale from pairwise Jev judgments. Updated 2026-10-12. - [laya-mlx runs the Laya checkpoints on Apple Silicon](https://jevainews.com/news/laya-mlx/): mizorewww/laya-mlx is an Apache 2.0 MLX port of Convai's Laya weights. Updated 2026-09-30. - [llamacpp-jev answers /v1/systemone from an unmodified llama-server, including images](https://jevainews.com/news/llamacpp-jev/): Chirag posted llamacpp-jev, an MIT wrapper that puts TypeSafe's decision API in front of stock llama-server. Updated 2026-10-05. - [pg-jev puts prompt_jev() in PostgreSQL and returns jsonb](https://jevainews.com/news/pg-jev/): Jose Mejias posted mejiasd3v/pg-jev, an MIT extension. prompt_jev sends text to Jev from SQL and returns a probability, a choice, or a score as jsonb. Updated 2026-10-03. - [A Hold'em table asks Jev about the opponent and lets code pick the bet](https://jevainews.com/news/poker-jev/): A September 22 post links Jinzhengxu/poker-table and says each action averaged 545 ms and a hand cost $0.002. Updated 2026-09-30. - [A chart puts Jev's reading of 25 probability words next to 99 people](https://jevainews.com/news/probability-words/): Mike Taylor posted a chart of 25 phrases from a 2026 survey of 99 people. Updated 2026-09-30. - [rev reads a frozen Qwen and posts 0.784 on public hard items](https://jevainews.com/news/rev-local/): Steven Yu posted 54yyyu/rev, an MIT harness that reads option logits from a frozen open model. Updated 2026-09-30. - [spark-s1-4b-v6 swaps in Qwen3.5 and prints a public-tier 83.1](https://jevainews.com/news/spark-s1-v6/): Abhishek Rai posted spark-s1-4b-v6, an Apache-2.0 LoRA on Qwen3.5-4B. Updated 2026-09-30. - [swift-jev types Choice, Noul, and Score, and routes at 0.60 and 0.85](https://jevainews.com/news/swift-jev/): Daiki Matsudate posted a Swift client after a trial that paired Jev with another model did not go well. Updated 2026-09-30. - [djev reads DiffusionGemma as a decision API, and sits third on JevBench](https://jevainews.com/news/djev-diffusion/): Florian S's JevBench table puts djev at 74.3, 1.1 behind Jev. Updated 2026-09-30. - [jevsearch re-ranks keyword hits with Jev and scored 83% Hit@1 on the author's TypeSafe docs set](https://jevainews.com/news/jevsearch/): Kyle McLaren posted jevsearch, an MIT shadcn registry block that streams keyword hits and then re-ranks them with Jev. Updated 2026-09-30. - [An n8n community node routes Jev answers onto category branches, including Needs Review](https://jevainews.com/news/n8n-jev-classification/): khmuhtadin posted n8n-nodes-jev-classification, an MIT community node that wraps TypeSafe's Jev as classify, score, check, and ask. Updated 2026-09-30. - [A 100-case security triage demo put Jev at 65.3% balanced accuracy against Terra 75.1% and Opus 91.2%](https://jevainews.com/news/security-triage/): RZ posted a recorded comparison of Jev, OpenAI Terra in Codex, and Claude Opus in Claude Code on 100 synthetic TypeScript security cases, five passes each. Updated 2026-10-07. - [SimpleJev turns an open model into a classifier by reading the next-token logits](https://jevainews.com/news/simple-jev/): Featherless posted SimpleJev, an Apache 2.0 server that scores Choice, Score, and Noul from a Hugging Face model's next-token logits. Updated 2026-09-30. - [Spring publishes a Java client that asks Jev before the chat model runs](https://jevainews.com/news/spring-ai-typesafe/): Opened October 8, 2026, the README quick start pins 0.4.0, and Maven Central lists that release. Updated 2026-10-08. - [AstroHan ran 30 coding tasks with a Jev context filter: 25 passed, 22 without it](https://jevainews.com/news/compaction-30-tasks/): AstroHan scored every tool result with Jev before it reached deepseek-flash, on the same 30 FrontierHarness tasks. Updated 2026-09-30. - [DocJev classifies and splits real PDFs with Jev, 5.7x to 6.5x faster than Luna on the author's set](https://jevainews.com/news/docjev-documents/): Jerry Liu posted DocJev, an Apache 2.0 library that classifies and splits PDFs with local LiteParse and hosted Jev. Updated 2026-09-30. - [Jared Palmer's Kev family copies Jev's API on Qwen, and trails it on new sources](https://jevainews.com/news/kev-family/): Cognition's VP of engineering published Kev, an Apache 2.0 family of Jev-like models you train and serve yourself. Updated 2026-10-02. - [Laya, an Apache 2.0 System One stack, publishes where it beats Jev's numbers and where it does not](https://jevainews.com/news/laya-open-weights/): Louis Grenard pointed at Laya, Convai Innovations' open typed-decision engine. Updated 2026-09-30. - [semgate asks Jev about each HTTP request, then a live range to attack it](https://jevainews.com/news/semgate-go/): mizutani published semgate, a Go net/http middleware that sends each request to Jev as typed questions and lets your function block, route, or pass. Updated 2026-09-30. - [GitHub Next ships LocalJev, a prompted stand-in that speaks /v1/systemone](https://jevainews.com/news/github-next-localjev/): GitHub Next published LocalJev, a Bun server that turns Jev questions into JSON-probability prompts for models running on oMLX. Updated 2026-09-30. - [LlamaIndex's Logan Markewich ships jeff, a local System One API on GLiFormer](https://jevainews.com/news/jeff-local-gliformer/): Logan Markewich released jeff, a self-hosted drop-in for TypeSafe's /v1/systemone endpoint backed by Knowledgator's GLiFormer. Updated 2026-09-30. - [Jev macOS Loop clicks native Mac windows; Finder sorted nine files in 7.39 seconds](https://jevainews.com/news/jev-macos-loop/): Josh C. Simmons open-sourced a computer-use loop for Apple silicon that keeps pixels on the Mac and asks Jev only over observed controls. Updated 2026-09-30. - [Jevis is a Flutter integration-test agent that only taps what you registered](https://jevainews.com/news/jevis-flutter/): Jake Gwon released jevis, a Dart package that drives Flutter's integration_test loop with Jev. Updated 2026-09-30. - [LangChain ran Jev as a judge: 500 matching pass/fail labels, and a $0.34 bill against Claude's $28](https://jevainews.com/news/langchain-jev-judge/): Daniel Shea and Seán Roche compared Jev with GPT-5.6 Luna, GPT-5.6 Terra, and Claude Sonnet 4.6 on five frozen weather-agent traces. Updated 2026-09-30. - [Jason Lemkin tried Jev on SaaStr Connect and stopped on a 32% false-admission rate](https://jevainews.com/news/saastr-false-admissions/): Jason Lemkin ran 600 judgments against Claude Sonnet. Updated 2026-09-30. - [SemIf, a 4B open rebuild that reads option logits, sits 0.7 behind Jev on JevBench](https://jevainews.com/news/semif-open-logits/): Theodore Lee's SemIf (formerly OpenJev) scores typed options from a frozen Qwen3.5-4B by reading logits, not by sampling JSON. Updated 2026-09-30. - [Sam Saffron tries Jev as a hidden control plane for GPT Live in term-llm](https://jevainews.com/news/term-llm-control-plane/): Discourse cofounder Sam Saffron opened a term-llm pull request that classifies each GPT Live voice turn with TypeSafe's System One API in about 650 milliseconds, instead of a tool-calling fast-model turn. Updated 2026-09-30. - [Aman Kumar ran about 16,000 Jev calls and called it a filter, not a classifier](https://jevainews.com/news/aman-kumar-filter/): Aman Kumar compared Jev with gpt-5.4-mini and gpt-5.6-luna on four public sets, then on production page gates and an inbox. Updated 2026-09-30. - [Bespoke Nimble 9B matches 90% of its own labels and 74.8% of a public suite](https://jevainews.com/news/bespoke-nimble/): Bespoke Nimble 9B scores 90.12% vs 93.21% for Jev 1.13.0 on 324 synthetic holdout labels. Updated 2026-10-06. - [Charly Poly put Jev next to open encoders, not GPT-5.6, and it lost on calibration](https://jevainews.com/news/charly-poly-encoders/): Browserbase's Charly Poly ran Banking77 against ModernBERT, BART-MNLI, and GLiClass. Updated 2026-09-30. - [Hono's author routes HTTP by meaning, then warns you not to](https://jevainews.com/news/hono-semantic-router/): Yusuke Wada published hono-jev-router, an experimental Hono router that asks Jev which plain-language description matches an incoming request. Updated 2026-09-30. - [Jev scored 384 morning headlines in 25 seconds; Opus 5 finished four](https://jevainews.com/news/jev-news-desk-dealer/): Elvis ran a morning Google News wire through Jev and Claude Opus 5 at the same time. Updated 2026-09-30. - [A hybrid Jev overlay sped up Postgres joins 12% on JOB, and made the planner slower](https://jevainews.com/news/jev-postgres-planner/): Michael Malis asked Jev to help plan Postgres joins on the join order benchmark. Updated 2026-09-30. - [SREGym put Jev on a Codex agent: 24 passes out of 50, up from 20](https://jevainews.com/news/jev-sregym-lite/): Jackson Clark's team wired Jev into SREGym-Lite as a test ranker and evidence reviewer on a Codex agent that still diagnosed and repaired. Updated 2026-09-30. - [Almeida's bet on Jevons paradox over chatbot price cuts](https://jevainews.com/news/jevons-paradox-bet/): In replies three days after launch, TypeSafe's founder argued that the business depends on enabling new software loops rather than capturing existing language model budgets. Updated 2026-09-30. - [TechCrunch covers Jev, and quotes a Vercel engineer who swapped Luna for it](https://jevainews.com/news/techcrunch-jev-developers/): On October 9, 2026, Marina Temkin's TechCrunch In Brief says TypeSafe raised $870 million at a $7.5 billion valuation. Updated 2026-10-11. - [Vercel's October 2026 AI Gateway Production Index prints Jev at 35.3% of paid teams in its first two weeks](https://jevainews.com/news/vercel-gateway-adoption/): Vercel's October 2026 AI Gateway Production Index, published October 15, prints Jev at 35.3% of paid teams in its first two weeks, nearly double Sonnet 5.5 at 18.1%. Updated 2026-10-17. - [A week of Jev repos: routers, MCP, graphs, games](https://jevainews.com/news/awesome-jev-week-one/): A widely shared Awesome Jev thread listed fourteen working patterns; community directories passed a hundred projects within days. Updated 2026-09-30. - [Browser Use wires Jev into the click loop: 7 seconds, $0.0039](https://jevainews.com/news/browser-use-jev-flights/): Gregor Zunic showed a 1x-speed Google Flights run where Jev picks the action and DOM target each step, and a small LLM types only when it has to. Updated 2026-09-30. - [Bryo ran 1,565 business emails: Jev at 96.4%, Gemini a point or two higher](https://jevainews.com/news/bryo-email-1565/): Nikhil Mudholkar, CTO of Bryo AI, routed 1,565 German and English supplier emails into 10 categories. Updated 2026-09-30. - [Compaction without summarization: score the tool calls and drop them](https://jevainews.com/news/instant-compaction/): Tamara Tran argued that agent context compaction is still a summarization prompt in 2026, and that Jev can instead keep or drop tool output verbatim. Updated 2026-09-30. - [LangChain ships a Jev classifier, then parks it on the agent loop](https://jevainews.com/news/langchain-jev-harness/): Official LangChain docs add langchain-typesafe, introducing a Runnable for classification alongside experimental middleware hooks for agent loops. Updated 2026-09-30. - [Stagehand, Chrome, and a11y trees: computer use as a Jev question](https://jevainews.com/news/stagehand-computer-use/): Kyle Jeong of Browserbase and other builders structured computer-use around structured observation, typed queries, and automated execution, reporting sub-cent task costs. Updated 2026-09-30. - [Tetris as a legal-move test: Jev picks from the landings you give it](https://jevainews.com/news/tetris-legal-moves/): Two independent demonstrations evaluated Jev playing Tetris from enumerated legal placements, comparing latency and costs against Claude Haiku 4.5 and Gemini 3.5 Flash-Lite. Updated 2026-09-30. - [TypeSafe starts pulling people off the Jev waitlist](https://jevainews.com/news/waitlist-opens/): TypeSafe began admitting waitlist users in batches, reporting 20,000 approvals in an hour while addressing browser compatibility issues. Updated 2026-09-30. - [One builder ran Jev on 1,500 of his own emails](https://jevainews.com/news/email-triage-1500/): Ryan Vogel posted a long clip of Jev classifying his mail. Updated 2026-09-30. - [A buy-or-sell every Monad block, with a mock default model](https://jevainews.com/news/jev-trader-monad/): Jarrod Watts demonstrated a trading bot querying Jev for buy or sell decisions on each Monad block. Updated 2026-09-30. - [JevPilot is a playable driving sim that asks Jev which path to take](https://jevainews.com/news/jevpilot-fsd/): Justin Schroeder wired Jev into a Three.js driving simulator in under an hour, then open-sourced JevPilot the next day. Updated 2026-09-30. - [Every ran 777 judgments in under a second and still wanted a better accuracy check](https://jevainews.com/news/every-777-judgments/): Mike Taylor at Every put 37 documents through 21 AI-writing questions at once. Updated 2026-09-30.