Jev changelog.
This is the permanent, dated ledger of material updates, model releases, provider integrations, and pricing adjustments for Jev.
Every entry in this ledger is verified against authoritative primary documentation. We do not publish speculative future roadmaps or unverified rumors. Current production baseline: jev-1.13.0.
Verified Event Timeline
Chronological (Most Recent First)TypeSafe & OpenRouter Launch Jev Router (typesafe/jev-router)
TypeSafe and OpenRouter launched 'typesafe/jev-router', a managed model router. Running on TypeSafe's Jev System One decision model, Jev Router picks the candidate model and reasoning effort for each request to balance quality, speed, and cost, returning text completions via OpenRouter's standard Chat Completions API.
- Exposes OpenAI-compatible POST 'https://openrouter.ai/api/v1/chat/completions' using model string 'typesafe/jev-router'.
- Supports up to 1,000,000 tokens of context across multimodal inputs (audio, file/PDF, image, text, video) and returns text.
- Dynamically determines model choice and reasoning effort parameters, surfacing the selected backing model in the response 'model' attribute.
- Currently listed at $0.00 prompt and completion token pricing on OpenRouter.
Primary source / OpenRouter model catalog & TypeSafe ↗
Vercel AI Gateway Adds TypeSafe Client Route & HTTP Evaluation API
Vercel expanded AI Gateway support for TypeSafe Jev ('typesafe-ai/jev'), adding native compatibility for the '@typesafe-ai/sdk' client library via 'https://ai-gateway.vercel.sh/typesafe' and a direct HTTP evaluation API at 'POST https://ai-gateway.vercel.sh/v1/evaluate' in addition to existing AI SDK access.
- Enabled '@typesafe-ai/sdk' usage through Vercel AI Gateway using baseURL 'https://ai-gateway.vercel.sh/typesafe' and 'AI_GATEWAY_API_KEY'.
- Added direct REST evaluation endpoint POST 'https://ai-gateway.vercel.sh/v1/evaluate' using model string 'typesafe-ai/jev'.
- Provides three distinct integration pathways alongside AI SDK 'experimental_evaluate'.
Primary source / Vercel AI Gateway documentation & changelog ↗
OpenRouter Adds Jev Support with Dual Routes
OpenRouter made Jev available to platform developers under model identifier 'typesafe/jev-1.13' and default alias '~typesafe/jev-latest'. OpenRouter lists a 32,000-token context window and matches TypeSafe's published token pricing ($0.042/M input tokens, free output tokens).
- Stable route launched at POST https://openrouter.ai/api/v1/systemone supporting full multi-question batching.
- Experimental single-decision interface exposed via OpenRouter Jev Lab.
Primary source / OpenRouter model directory ↗
Vercel AI Gateway Adds Jev Integration
Vercel AI Gateway announced initial provider routing for TypeSafe Jev ('typesafe-ai/jev'). AI SDK exposes Jev through an experimental evaluate API, supporting low-latency decision routing and pass-through key management.
- Exposes Choice, Score, and Boolean evaluation via the AI SDK experimental_evaluate interface.
- Pass-through authentication to TypeSafe cluster endpoints with zero data retention options.
Primary source / Vercel AI Gateway changelog ↗
TypeSafe AI Launches Jev 1.13 in Early Access
TypeSafe AI publicly unveiled Jev version 1.13 (versioned ID: 'jev-1.13.0', alias 'jev-latest') as a non-autoregressive System One decision model trained via Reinforcement Learning for Calibrated Decisions (RLCD).
- Introduced three core evaluation primitives: Choice (discrete selection ≤255 choices), Score (ordered rubrics with 2–10 levels returning continuous weighted floats), and Noul (binary proposition truth probability float).
- Established dual context limits: 64,000 tokens total request budget across state + all questions, with a 32,000-token cap for state + single longest question.
- Set launch pricing at $0.042 per 1,000,000 input tokens with $0.00 output token billing.
- Published official SDK packages: 'typesafe_sdk' for Python and '@typesafe-ai/sdk' for TypeScript/JavaScript.
Primary source / TypeSafe launch announcement by CEO Diogo Almeida ↗
How this ledger is maintained
Entries are recorded only after public release by TypeSafe AI or authoritative third-party platform providers. Unannounced leaks, unsubstantiated benchmarks, or speculative release timelines are strictly excluded from this ledger.