THE JEV GUIDE · CURATED EDITION
Launch news, explanations, demos, and community projects around TypeSafe Jev, organized by topic. Every entry links back to its X source.
Curated selection · Not exhaustive · Updated manually
01 / EXPLORE
0 means awaiting a verified post
The post says Jev solved a problem AI teams have patched for years: parsing LLM JSON output, handling markdown fences, and dealing with inconsistent enum values.
TypeLLM is a new open-source model that reasons before producing type-safe output, dramatically boosting accuracy without fine-tuning. The team claims it surpasses Jev and GPT-5.6 Luna and approaches GPT-6 Astra. Link
As the first System One model from TypeSafe AI, Jev is built for fast, structured decisions inside software. Instead of generating text, it takes app state plus a typed question and returns a typed decision with probability. Link
Announces Jev-Omni, a model built on Gemma 4, aimed at challenging Jev.
TypeSafe AI's first System One model, Jev, is now available on the B.AI API. Designed for high-speed structured decision-making inside software, Jev does not generate text. Instead, it takes application state and typed questions as input and directly returns decisions with probabilities and confidence scores, supporting Choice, Score, Noul, and more. Link
The post mentions TypeSafe's launch of Jev, which classifies texts among preset answers. The author mocks the hype, noting the format is guaranteed but not the correct answer, questioning how naive buyers are assumed to be.
TypeSafe AI introduces Jev, a new AI model that focuses on making structured decisions from context rather than generating text.
Scoreboar v8 reduces memory from 597 MB to 128 MB and per-post processing from about 1 second to about 30 ms, while improving accuracy. It guesses which of two X posts performed better for its account size with 61.4% accuracy, beating Grok 4.7 at 55.3% and Jev at 52.5%. Free and runs in your browser. Link
OMH adds support for Jev skills and model routing, including Jev models and aliases, skills and plugins, workflow routing, fast decisions, plus plugin status, credentials, and conflict checks.
TypeSafe plans to unveil Jev on Monday. Jev does not generate text; it only returns answers from the schema based on state and typed questions, taking 70-500ms. Input costs $0.042/M tokens, output is free, and access is waitlist-based. Demos include Mario, real-time levels, and color selection. It cannot return content outside the schema.
JEVTOWN is an AI social simulation platform, similar to the Lobster agent forum, where 10,000 AI personas browse, like, retweet, block, or buy, with each reaction decided by Jev. It has reached #16 on Product Hunt. Link
Jev is an AI model built purely for fast decisions, not conversation. It returns only the chosen answer and the probability it's correct—no token-by-token generation, just a decision in a single parallel pass.