InfoQ covers Jev's release: Vercel uptake, a 7× median speedup and the jaggedness caveat

Source · Model card: Jev

InfoQ (“TypeSafe AI Releases Jev: A Decision-Only Model That Returns Typed Probabilities Instead of Text”, by Daniel Curtis, 1 October 2026) pulls together what has been reported about Jev since launch.

Recap. $0.042 per million input tokens with free output, a 32,000-token context, 70–500 ms end-to-end latency quoted by TypeSafe, and training with Reinforcement Learning for Calibrated Decisions (RLCD).

Adoption and third-party numbers (UnverifiedClaim).

• Vercel added Jev to AI Gateway on day two and says it reached nearly 13% of paid teams within 24 hours, twice the share of the GPT-5.6 family. Netlify followed; LangChain shipped a TypeSafeClassifier integration and an AutoMode middleware that screens tool calls; five independent Elixir clients appeared.
• A Vercel engineer measured a safety classifier running 5–18× faster than the LLM it replaced; Bryo AI’s CTO found Gemini slightly more accurate on email classification but 10–20× more expensive.
• OpenChamber analysed 12,759 launch tweets: user-reported speedups at a median of 7× against the 193.6× headline, cost savings at a median of 30×, latency at a median of 76 ms (upper quartile 270 ms).
• Armin Ronacher (Earendil), quoted from TechCrunch: the design “delegates the hallucination problem a little bit to the user”.

Practical advice in the piece. Start with the jev-1.13 jaggedness page (counting, arithmetic, date comparison and large noisy state are weak; keep maths in code), pin jev-1.13.0 instead of the moving aliases, and use the System One adapter to benchmark existing models on the same schema.

Why it matters. The OpenChamber median is a useful outside correction to the launch speedup. Jev’s scores are unchanged (6.0 / 8.2 / 58); the production signals will be weighed at the next adoption review.