Typesafe AI
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 15
- Integrating the Jev judgment model with Codex
Juejin author 二哥呀 demonstrates how to install TypeSafe AI’s Jev Skill in Codex, configure an API key, and require the agent to call Jev’s Choice API before taking an action. In the shown example, Jev 1.13.0 selected web search for an Nvidia stock-price question with 0.99 confidence, after which Codex executed the selected action. The author also says Jev works with Claude Code and PI, while speed and cost comparisons come from TypeSafe AI rather than independent testing.
掘金 · 人工智能 · 🔥 0 - Cloudflare open-sources Clef decision models for AI agents
Cloudflare released Clef, a family of open-weight models that pick among predefined options instead of generating text, in 27B-parameter and 9B-parameter (Clef-Flash) versions, InfoQ reports. The 27B multimodal model takes a state plus a schema of typed questions, handles text, JSON, images or video, and returns probabilities for all options in a single forward pass, while Clef-Flash targets latency-sensitive decisions; Cloudflare's own benchmark puts Clef-Flash at 38.8 ms median latency versus 209.3 ms for the 27B model. Weights are downloadable on Hugging Face and the models are served through Workers AI, alongside a fine-tuning service staffed initially by Cloudflare engineers, and the API is compatible with Typesafe AI's Jev System One, though Hacker News and Reddit commenters questioned public benchmarks and whether the models know when to abstain.
InfoQ 中文 AI&大模型 · 🔥 0 - China Telecom's Xingchen lab open-sources Xing4.0-29B-Intent intent-parsing model
China Telecom AI's Xingchen AGI lab released and open-sourced Xing4.0-29B-Intent on October 19, a 29B model that combines intent classification with parameter extraction and returns results as schema-constrained JSON. Alongside the weights it published 2.753 million supervised fine-tuning examples rebuilt from 61 public datasets covering 19 domains and 52 languages, plus an inference SDK and official images under Apache 2.0. Zixun ran 11 hands-on tests across smart spaces, smart cockpits, office collaboration and personal-agent scenarios and reported generally accurate multi-object parameter binding; the headline benchmark numbers, including an overall score of 58 against 49 for Gemini-3.6-Flash across 48 public datasets, come from the model team's own evaluation, which the reports note does not represent all tasks or deployments.
智东西 · 🔥 0 - Jev becomes Vercel's fastest-adopted AI model ever
Vercel data published on Thursday shows that TypeSafe AI's lightweight decision model Jev was used by 35% of paid teams within two weeks of its limited release, nearly double the reach of the second most popular model, Sonnet 5.5, making it the fastest-adopted model in its first two weeks on Vercel's platform. Released in limited form on September 15, Jev restricts outputs to a few predetermined options to run cheaper and faster, making it suited to classifying large volumes of data; Vercel has used it to replace GPT-6 Luna for classifying harmful commands. The figures come from Vercel's AI gateway, a unified API service similar to OpenRouter, and may undercount labs such as OpenAI and Anthropic that offer their own APIs; the same report says open-source models account for more than 70% of tokens processed through the gateway and nearly 25% of total spend.
TechCrunch AI · 🔥 0 - Liquid AI releases d1-ultra decision model
Liquid AI announced d1-ultra, a multimodal decision model it says is built for harder real-world cases such as spotting a defect on a circuit board against its spec or checking a damage claim against a customer photo. The model takes text, images or both and returns probabilities for yes/no, choice and score questions in one forward pass without generating tokens. It is available on the Liquid API at $0.042 per million input tokens; Liquid AI said it would soon be available through Vercel and OpenRouter, and a Liquid AI post said it was now available on OpenRouter. Liquid AI says d1-ultra outperformed Jev 1.14 across all tasks in a comparison spanning 112 evaluations, and that it leads all other decision models on the Artificial Analysis Decision Index by at least 8 points. IT之家 reported that TypeSafe AI released the Jev 1.14 preview a day before Liquid AI announced d1-ultra, and that d1-ultra performed better than Jev 1.14 in eight task categories and better overall than competitors such as vLLM Semantic Router d3.
Liquid AI Blog · 🔥 0 - TypeSafe ships Jev 1.14.0 with image input, Box grounding and 262k context
TypeSafe AI released Jev 1.14.0, which the company describes as an unflashy, small update with broad but subtle changes and the clearest expected gains in multilingual performance; it warns that non-English languages, Chinese included, still lag English. The release also opens three experimental features grouped under a "Danger Zone" program: image input, a new Box question type that returns normalized bounding-box coordinates, and long context that raises per-request and per-question limits to 262k tokens. All three are off by default and require manual API switches, and TypeSafe says they do not meet production standards and are used at the developer's own risk. Pricing is unchanged at $0.042 per million input tokens with free output, and TypeSafe's internal evaluation put Jev 1.14.0 first among eight models at 60.0 while cautioning the benchmark may favor its own model and scores may change. DoNews also reported that Liquid AI launched its d1-ultra decision model the next day, claiming it beats Jev 1.14 across eight task types, a vendor claim.
智东西 · 🔥 0 - DecisionRobust benchmark tests JEV-style decision models against text and image attacks
A new arXiv paper introduces DecisionRobust, a benchmark for systematically evaluating how robust JEV-style decision models are to textual and visual inputs. It covers practical decision-making scenarios with semantics-preserving text perturbations, image perturbations and prompt injection in both modalities. The authors evaluated 30 model configurations on text tasks and 17 on vision tasks, including 24 and 12 held-out models respectively; they report that attacks expose vulnerabilities beyond the models used for construction and selection, that optimized image perturbations transfer weakly to many targets, and that resistance to input perturbations does not necessarily imply resistance to prompt injection.
Hugging Face · Papers · 🔥 0 - vLLM Semantic Router open-sources Decision 3.0 decision models, tops decision leaderboard
The vLLM community project vLLM Semantic Router released and open-sourced the Decision 3.0 family of multimodal decision models: d3, d3-flash, d3-mini, d3-nano, d3-lite and d3-edge, spanning roughly 0.6B to 27B parameters, with weights published on Hugging Face. The flagship d3 is built on Qwen3.8-27B and accepts multiple images and video, while the other five are based on Qwen3.5 at different sizes; unlike Jev, which is served mainly via cloud API, the series is open-weight and locally deployable. According to team-provided figures, on the Jev Decision Index 0.3.1 leaderboard d3 ranks first with a text decision index of 64.7 and a vision score of 71.6, ahead of Jev's 60.1 and Microsoft Decision-1's 53.2 (9B, fine-tuned from Qwen3.5-9B); the team also reports median latency on a single AMD Instinct MI325X GPU of 55 ms for text, 273 ms with images and 553 ms for 10-second video, while one developer ran d3-lite via Core ML on an M5 Pro Mac to classify and route 1,000 pull requests in about two minutes.
智东西 · 🔥 0 - Microsoft launches Decision-1, a Qwen-based model that scores fixed choices
Microsoft released Microsoft-Decision-1, a small model that does not write open-ended text but returns a calibrated probability for each of a fixed set of options, aimed at routing, classification, prioritization, verification, workflow and agent control. According to the Foundry model card, it was built by post-training Alibaba's open-weight Qwen3.5-9B; Microsoft software engineering VP Achint Srivastava announced it in a 9 October blog post and said the company will soon rebase it on other models, including its own MAI models and OpenAI's. It is available in Microsoft Foundry and due on OpenRouter, priced at $0.042 per million input tokens with output tokens free. Microsoft says it posted the highest accuracy, 83.5%, across a 36-benchmark comparison covering nearly 150,000 questions held out from training, and claims it is 2.5 times faster than runner-up H2O-Lightning-4B, 35 times faster than GPT-6 Sol and 2.8 times faster than Jev — all vendor-run comparisons.
TestingCatalog · 🔥 0 - Jev maker TypeSafe AI raises $870M at $7.5B valuation led by a16z
TypeSafe AI, the startup behind the Jev model, said it raised about $870 million in a Series A led by Andreessen Horowitz, with Sequoia Capital and existing investor DCVC participating, at a roughly $7.5 billion valuation; Martin Casado is joining the board. The company describes Jev, which went viral after its September 15 release, as built on a transformer architecture but not a large language model: instead of producing text it outputs probabilities, which TypeSafe AI calls “calibrated decisions,” and the company claims it runs faster and uses far fewer tokens than LLMs. TypeSafe AI also says about a third of Fortune 500 companies are already using Jev, a figure it has not detailed or independently verified.
Bloomberg Technology · 🔥 0 - Jev: TypeSafe AI's probability-only model, explained
TypeSafe AI emerged from stealth on September 15 with Jev, a model that does not produce free-form text but instead returns estimated probabilities over a fixed answer set — yes/no, multiple choice, or a rating. It went viral: the announcement drew nearly 40 million views on Twitter, over 100,000 people joined its Discord within a week, some 2,000 GitHub projects adopted it, and signups opened on September 20 were paused two days later amid overwhelming demand. Because it is faster and cheaper and its output slots directly into code patterns like if statements, Jev has spawned similar models from the likes of Cloudflare, Amazon and OpenAI.
Understanding AI · 🔥 0 - AI's simmering question: where is all the automation?
A The Information column frames the industry debate over why automation has not arrived even as OpenAI said its unreleased model solved or made major progress on hundreds of significant math and computer science problems. Cognition founder Scott Wu argued that ever-more-capable agents plus more data centers, memory and sandboxes will yield "virtual employees," while ex-OpenAI researcher Diogo Almeida, now at TypeSafe AI, maker of Jev, said today's LLMs are optimized for assistance and fail at unattended automation. Anthropic's Claude Code product lead Cat Wu conceded much automation stalls at a semi-automated stage where Claude does 80% and a human does 20%.
The Information · 🔥 0 - OpenAI launches Decisions API for fast choices, targeting Jev
At DevDay, OpenAI announced a Decisions API that narrows open-ended generation to picking from predefined options and returns a choice plus confidence that code can consume directly, which The Decoder says runs about ten times faster than the Responses API. It currently supports only GPT-6 Luna, at $0.10 per million input tokens with free output tokens in public beta, according to The Decoder, while InfoQ describes it as a limited preview with pricing and customization details to come at general availability. OpenAI also cut its paid API tiers from five to three: Build, Launch and Grow.
The Decoder · 🔥 0 - TypeSafe AI’s Jev turns frequent AI tasks into fast decisions
TypeSafe AI’s Jev is a decision model designed to return structured choices, scores, or calibrated yes/no probabilities instead of conversational text. The model launched on September 15 and drew developer attention for reported low latency and pricing, while TypeSafe AI claims of large speed and cost gains have not been independently verified.
创业邦 科技 · 🔥 0 - TypeSafe AI’s Jev sparks interest in non-generative decision models
TypeSafe AI’s Jev is a decision model that maps inputs to predefined outputs such as yes/no answers, scores, or list selections instead of generating free-form text. CEO Diogo Almeida claims the model is used by about 25% of Fortune 500 companies and processed one trillion tokens per day roughly a week before the report, while separate reports say the startup is discussing a new funding round; these claims were not independently verified in the source.
IT之家 AI · 🔥 0
Experience and discussion from the community
Share my Typesafe AI experienceAsk about Typesafe AI
Nobody has shared their experience with Typesafe AI yet.