Jev getting-started guide: a decision-only System One model
An introduction to jev-1.13.0's Choice, Score, and Noul question types, its $0.042 per million input tokens with free output, and two-minute steps to wire it into an agent loop.
holy sh*t, it can't be f**king real TypeSafe just dropped Jev a decision-only model with free output and $5 credit to start Jev is the first System One model. it doesn't write, it only decides. you send it state + typed questions, it returns decisions with calibrated probabilities. no prose, no parsing, no second-guessing. put it between your generation model and your actions route, rank, verify, escalate, loop-health without burning frontier tokens on every gate what you get: - Jev (jev-1.13.0) no text generation, only decisions - three question types: Choice, Score, Noul - output completely free, input at $0.042 per million tokens - $5 free credit to start covers over 100K decisions - 64K context, evaluates every question in parallel against the same state how to try (under 2 min): step 1: go to https://typesafe.ai step 2: request API access (waitlist) or use OpenRouter / Cloudflare / Vercel AI Gateway same $0.042 pricing, no waitlist step 3: send state + questions to POST https://api.typesafe.ai/v1/systemone model id: jev-1.13.0 step 4: wire it into your agent loop generator writes → Jev decides → code branches. that's the whole frame important: > Jev cannot generate text it's not a chat model, don't point an OpenAI client at it > access is waitlist-gated direct from TypeSafe, but gateways serve it today > per TypeSafe, probabilities are calibrated across groups, validate on your own data > $5 credit is for direct TypeSafe access only
This post quotes another post: x.com/painn_x/status/2102033746480304364