Typesafeai · since 15. September 2026 (early access) · 2× · last seen Sep 22, 2026
Jev is TypeSafe AI's first public "System One Model" (founded by ex-OpenAI researcher Diogo Almeida), unveiled in early access on September 15, 2026. Unlike conventional LLMs, Jev does not generate free text; instead it takes a "state" plus typed questions (Choice, Score, Noul) and returns typed decisions with calibrated probabilities and confidence scores in a single parallel pass rather than token-by-token. TypeSafe claims Jev is up to 100–200x faster and cheaper than comparable LLMs on classification/decision tasks, though these figures are self-reported and not yet independently verified. The model is only available as a hosted, closed-weight API; architecture, weights, and parameter count have not been disclosed.
Features
| Deployment (Self-Hosted/Cloud) | Hosted cloud API only; no self-hosting, no open weights available |
| Throughput/Latency | Response time approx. 70–500 ms per call (typically ~100 ms); up to 193.6x faster than frontier LLMs per vendor claims |
| License | Proprietary, closed-weight; no weights or architecture published |
| Platform | Hosted cloud API by TypeSafe AI, San Francisco; also available via third parties like OpenRouter, Requesty, Cloudflare AI, Vercel AI Gateway, Opper |
| Price | $0.042 per 1M input tokens ($42 per billion); output tokens free |
| Protocol Compatibility | Not OpenAI chat-API compatible; dedicated POST /v1/systemone endpoint with state+questions body instead of messages/chat.completions |
| Release Date | September 15, 2026 (early access), alongside $40M seed round led by DCVC |
| Supported Models/Providers | Model version jev-1.13.0 (aliases jev-latest, jev-preview); one provider: TypeSafe AI, Inc. |