TypeSafe AI’s Jev model draws developer interest

Two years after leaving OpenAI, former researcher Diogo Almeida has released Jev through his startup, TypeSafe AI. The transformer-based model is not a large language model (LLM): instead of generating text, it produces probabilities, which the company calls “calibrated decisions.”

TypeSafe AI says Jev is cheap and fast because it avoids language generation. Users define its outputs in advance, which the company says prevents hallucinations. Output tokens are free, while input tokens are metered by the billion rather than the million. Demand briefly overwhelmed Jev’s API.

Developers are testing Jev for software automation. Vercel replaced OpenAI’s ChatGPT Luna 5.6 with Jev for a command-safety classifier and reported results 5 to 18 times faster, with greater accuracy. Bryo AI tested Jev against Gemini on business-email classification. Gemini was slightly more accurate, but 10 to 20 times more expensive.

Jev can also check other models, monitor LLM agent traces, help prevent jailbreaks, and route workloads to suitable models. Earendil CTO Armin Ronacher said its confidence scores could help users distinguish a 50% probability from a 95% probability.

Almeida says Jev is trained exclusively on synthetic data using “reinforcement learning from calibrated decisions.” The company calls it a “System One model,” while outside observers suspect it may build on an open-weight LLM. Named after 19th-century economist William Stanley Jevons, Jev is TypeSafe AI’s first model of this kind, though competitors may follow.