Jev, the Text-Free 'System One' Model: 200x Faster, ~400x Cheaper Than LLM Workflows
AI寒武纪 · wechat · 2026-09-22
Jev from TypeSafe AI has gone viral among developers and is now generally available. It's a "System One" semantic decision engine with no text-generation ability: given a state and typed questions, it returns strongly-typed choices with calibrated confidence — engineers call it "a switch statement with semantic understanding."
Key facts:
- Three output types: Choice (option + probability distribution), Score (ordinal levels), Noul (boolean probability). Output never violates the schema.
- Pricing: $0.042 per million input tokens, output free; 70–500ms latency; official benchmarks show 200x faster execution and 400x lower cost vs. generative-LLM workflows.
- Trained with RLCD (reinforcement learning from calibrated decisions): a stated 90% probability converges to 90% real accuracy. "No hallucination" only guarantees format compliance — the judgment itself can still be wrong.
- Best-fit uses: model routing, tool-call risk gating for agents (e.g., flagging destructive shell commands), and verification of agent outputs. Not for open-ended generation, deterministic computation, or long-chain reasoning.
- Adoption advice: replace your most brittle regex rule or a wasteful yes/no LLM call first, run in shadow mode to calibrate confidence thresholds, then switch traffic. Sign-up includes $5 credit (120M input tokens).
More from coding & agent
- Loom founder launches Roma 2.0, a shared to-do list that orchestrates AI agents — thetripathi58 · 2026-11-12
- Sosumi Eval: Apple Docs Access Lifts Coding Agent Pass Rates from ~33% to 94-98% — JordanMorgan10 · 2026-11-12
- CypherGuard fixes structurally wrong LLM Cypher queries, hitting 88.5% on CypherBench — JeremyCMorgan · 2026-11-12
- Building an End-to-End Hedge Fund in Python with Claude Code: A 5-Step Workflow — mdancho84 · 2026-11-12
- Jeff Dean: comfortable with engineers spending up to $10M/year on AI tools — deedydas · 2026-11-12
- Developer builds VladOS, a personal life dashboard, by just describing it to an AI agent — haltakov · 2026-11-12