Jev News & Timeline

This is a living page. We track Jev's early-access rollout and update this timeline as new information surfaces. Each entry names its source; anything we could not verify is marked as such. If jev.com disagrees with anything here, jev.com wins — tell us via /contact.

#Timeline

#2026-09-15 — Early access opens

TypeSafe AI opens early access to Jev, a judgment-only ("System One") AI model: structured state in, typed verdicts out (yes/no probabilities, multi-choice, scores) — no text generation. API keys via console.typesafe.ai.

Sources: SHISHIGAMI TECH (「9月15日に早期アクセス」); Greg Isenberg, "Jev is HERE. How to use it"; learnjev.com /start.

#2026-09 — Current model: jev-1.13.0

The production model is Jev 1.13 (jev-1.13.0), with aliases jev-latest and jev-preview. Endpoint: POST https://api.typesafe.ai/v1/systemone. Official SDKs: typesafe-sdk (Python 3.10+) and @typesafe-ai/sdk (Node 20+), both with automatic 429 backoff.

Sources: jevai.wiki /models/ and /api/; DNS cross-verification of typesafe.ai subdomains (2026-09-20).

#2026-09 — Official demo: Doom, played by judgments

TypeSafe AI's flagship demo runs Jev as the decision layer in Doom: game state goes in as structured input, movement/combat decisions come out as verdicts, in real time. It is the canonical illustration of the pitch — judgment fast and cheap enough to sit inside a game loop.

Sources: official demo coverage in early-access reporting; Moritz, "Jev: The New AI Model That's Breaking The Internet (Full Tutorial)" (demo walkthrough chapters).

#2026-09 — Pricing reported: $0.042/M input, output free

Community documentation reports pricing of $0.042 per million input tokens ($42 per billion), output tokens free; 64k context (state + longest question capped at 32k). Rate limits reported at 250k tokens/second and 1,200 requests/minute.

Sources: learnjev.com /tutorials/cost-and-benchmarks; jevai.wiki /models/. Early-access pricing — verify at jev.com before budgeting.

#2026-09 — ⚠ Rate limits may shift dynamically

Official materials warn that rate limits are subject to dynamic adjustment during early access. Design for 429s: queue, back off (SDKs do this automatically), and don't hard-code today's limits into capacity plans.

Sources: learnjev.com cost/benchmarks tutorial; jevai.wiki /api/.

#2026-09-20 — ⚠ Accuracy caveats: measure before you trust

Independent early testers report accuracy below official claims, varying widely by task; one third-party benchmark was published and later retracted. Our standing policy: official performance numbers appear on this site only with caveats attached, and every pipeline guide tells you to evaluate on your own data first.

Sources: learnjev.com critical coverage ("official numbers drift", retracted benchmark); cross-checked against community testing reports (2026-09-20).

#What we're watching

• Official documentation beyond the launch blog (our access to jev.com is intermittent — if you can reach it, send us corrections)
• Additional model versions beyond 1.13.x
• Whether early-access pricing survives general availability
• Non-text input support (images/audio currently require structuring first)

#Sources

• learnjev.com — /start, /tutorials/cost-and-benchmarks (dates, pricing, rate-limit warnings, accuracy caveats)
• jevai.wiki — /models/, /api/ (model versions, endpoint, SDKs)
• jev101.com — /zh/guide/what-is-jev.html (launch coverage, Chinese community)
• Early-access video coverage: Greg Isenberg, Moritz, Jack Roberts, SHISHIGAMI TECH, Caleb Writes Code, AI with Surya (all September 2026)

Unofficial fan-made handbook. Not affiliated with TypeSafe AI or jev.com.