Jev
The original 'System One' decision model — typed, calibrated answers at $0.042 per million tokens.
About
Jev, launched in early access on September 15, 2026, is TypeSafe AI's 'System One' decision model — the product that named the category. You hand it a state (text or JSON) plus questions defined in code, and it returns choices, scores, or yes/no probabilities, with no prose, no rationale, and no code. It is built for software, not people: the fast judgment a pipeline needs thousands of times an hour.
TypeSafe is a San Francisco lab founded in 2024 by Diogo Almeida (ex-OpenAI, co-author of the InstructGPT paper), Erik Gafni, and Sasha Sheng; it announced Jev alongside a $40M seed round led by DCVC, and raised a reported $870M in October 2026. The model is trained with Reinforcement Learning for Calibrated Decisions (RLCD), which rewards calibrated probability estimates rather than human preference — so a 0.8 confidence should mean roughly 80% observed accuracy, letting code branch on the numbers.
Pricing is $0.042 per million input tokens with output free, on a ~64K context. TypeSafe's homepage claims 193.6x faster and 444.6x cheaper — its best case against its slowest, priciest baselines; independent reruns measure roughly 3–5x faster and 40x cheaper than Claude Sonnet 5 at equal accuracy. Vercel reported 13% of its paid AI Gateway users calling Jev within 24 hours of launch. There are no open weights, and TypeSafe paused new signups about a week after launch on demand. The name honors economist William Stanley Jevons and his paradox: cheaper decisions mean more of them.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-claimed: 193.6x faster / 444.6x cheaper (best-case baselines). Independent reruns: ~3–5x faster, ~40x cheaper than Claude Sonnet 5 at equal accuracy; median 420–480 ms. Third-party bench (Oct 10, 2026): 79/82 correct, 0.450 s median.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Release facts (September 15, 2026 early access; $0.042/M input; RLCD training; founders and funding) corroborated across TypeSafe's announcement and multiple independent outlets; the 3–5x/40x figures come from independent reruns, not the vendor. Docked for: (1) the headline 193.6x/444.6x claim is TypeSafe's best case and misleading; (2) no open weights and no published parameter count; (3) new signups paused, so availability is constrained; (4) 'zero hallucinations' means schema-guaranteed, not correct — wrong valid values are still possible.