Ex-OpenAI researcher launches TypeSafe's Jev: System One models 100x faster, no hallucination
rohanpaul_ai · x · 2026-09-16
TypeSafe AI, founded by ex-OpenAI researcher Diogo Almeida, unveils its first "System One" model Jev after two years in stealth — a new class of frontier models for fast, structured decisions software can consume directly.
- New architecture + parallel sampler + a training method called Reinforcement Learning for Calibrated Decisions (RLCD)
- Matches existing LLMs' intelligence on System One tasks while being two orders of magnitude faster and more efficient
- Gives up string generation entirely; optimized for structured outputs and claimed to be incapable of hallucinating
- Positioned as "a frontier-intelligence function call": unstructured state in, typed probabilistic decisions out
- Available today in early access
More from Models
- TypeSafe Launches Jev: A Probability-Judging Model 25x Faster, 600x Cheaper Than Fable-Class Judges — soumitrashukla9 · 2026-09-16
- Local Flash Next beats Gemini Flash 3.8 on web game dev: 2 days vs 5 minutes, far better output — Forward_Jackfruit813 · 2026-09-16
- OpenAI's Astra to Surpass Fable 5.1 Soon, Dev Claims OpenAI Is 'Totally Back' — bindureddy · 2026-09-16
- OpenAI to retire GPT-5.5 from ChatGPT and Codex on Oct 14, but it stays on the API — OpenAIDevs · 2026-09-16
- Multiverse claims first AI model trained on quantum-generated data, mocked as hype — teortaxesTex · 2026-09-16
- Qwen3.8-27B-GSQ-RCO quant scores community-record 88.84/100 on 16 GB VRAM at ~31.7 tok/s — DerTomsn · 2026-09-16