Ex-OpenAI researcher launches Jev, an automation-first model claiming 200x speed and 100x cheaper
AI寒武纪 · wechat · 2026-09-16
TypeSafe AI, founded by ex-OpenAI researcher Diogo Almeida, has launched Jev, the first "System One" model designed for automation rather than chat.
- Approach: Instead of token-by-token generation, Jev outputs strongly-typed structured decisions with calibrated confidence probabilities, trained with a bespoke RLCD algorithm optimizing decision calibration.
- Performance and cost: 70–500ms end-to-end latency (40–200x faster), $0.042 per million input tokens with free output (100–1000x cheaper), and 0% structured-output/tool-call error rates by construction.
- Benchmarks: 68% accuracy on 4 production business workflows at the Pareto frontier, plus demos playing Doom ($7/hour) and Wikiracing. The team admits the headline speedup numbers are upper bounds from internally written workflows.
- Positioning: Best suited for classification, routing, scoring, Map-Reduce analysis, and guardrails embedded in regular software. Early preview is open.
Related event: TypeSafe exits stealth with decision model Jev and RLCD training method(86 posts)→
More from Models
- Anthropic's server-side rule reignites the case for owning open models over renting AI — alex_verem · 2026-11-02
- Epoch AI: open-weight GLM 5.3 exploited the patch gap to hack 6 Firefox releases at ~$1,000 each — rohanpaul_ai · 2026-11-02
- SemiAnalysis: safety classifiers sandbag scores of Opus 5.5 and GPT-6 Astra — morqon · 2026-11-02
- Liquid AI's 1.7B LFM3 becomes first local model to nail live fixtures lookup on a MacBook — NielsRogge · 2026-11-02
- Claude upgrade-page traffic rebounded to 17.6M in September, near May's 18.6M peak — firstadopter · 2026-11-02
- Open-source GLiNER-PII scrubs 13 identifiers from a discharge note in 550ms on GPU — MaziyarPanahi · 2026-11-02