Updates
Entries link to their sources. Dates identify individual updates; historical entries describe what was known at that time. Vendor measurements remain labeled.
2026-10-20
Updated SDK guidance for Python 0.7.5 and JavaScript 0.7.0, added model migration and experimental image guides, and corrected content dates, page titles, sitemap duplicates, and RSS identifiers. Examples distinguish illustrative output from live measurements.
Source: Jev AI Hub
2026-10-16
Updated guides and pricing config for jev-1.14.0 and the jev-preview alias, documented long-context headers and per-organization rate limits, refreshed GitHub star counts for published ecosystem projects, expanded the sitemap with /pricing/, /api/, and /getting-started/, and added a default Open Graph image plus an optional IndexNow postbuild hook.
Source: Jev AI Hub
2026-10-14
TypeSafe published jev-1.14.0 on the models page. Input pricing matches jev-1.13.0 ($0.042 per million input tokens, output free). jev-latest still points at jev-1.13.0; jev-preview now points at jev-1.14.0. Long context (262k limits) is available on both versions via the TypeSafe-Danger-Zone-Long-Context header. Public rate-limit numbers were removed in favor of per-organization limits in the developer console.
Source: TypeSafe Models
2026-09-30
New profiles for Kev, Laya, Von, SemIf, AnyJev, NanoJev, jevc, pi-jev-auto-mode, and jevmetrics, plus guides for the open-source landscape, coding-agent gates, confidence routing, composite scoring, and the official pattern and cookbook index. GitHub star counts were captured from the GitHub API on this date. Author-reported accuracy stays labeled as author-reported.
Source: Jev AI Hub
2026-09-30
The official models page still lists jev-1.13.0 at $0.042 per million input tokens, with output tokens free. jev-latest and jev-preview both still point at jev-1.13.0. There is no model named Jev 1.3. The same page now lists 100,000 tokens per second and 40 requests per second, and says those limits can move while demand is high.
Source: TypeSafe Models
2026-09-26
typesafe-sdk 0.7.2 adds a typesafe-sdk[http2] extra and documents HTTP/2 usage. The changelog lists no other code change. The JavaScript changelog still ends at @typesafe-ai/sdk 0.6.0.
Source: TypeSafe Python SDK changelog
2026-09-24
TypeSafe said on X that new direct console signups are closed for now, with no reopening date in that post. Existing accounts were not described as affected. Hosted gateways such as OpenRouter and Vercel were described as still available.
Source: TypeSafe on X
2026-09-24
Four guides on browser-use/jev-ultrafast: the 7.073-second Flights recording and its n=3 caveat, the snapshot and speculative fan-out loop, Jev as an agent decision layer, and a tutorial for the same contract.
Source: Jev AI Hub
2026-09-22
docs.typesafe.ai/cookbooks lists worked recipes in five groups: self-consistency, batching, how-to, extraction, and classification. The page assumes you already know the primitives and how confidence works.
Source: TypeSafe Cookbooks
2026-09-21
TypeSafe published “Jev with coding agents.” It says Jev is not a drop-in replacement for the LLM behind Claude Code, Cursor, Copilot, or similar tools. The recommended path is to keep that agent and have it write code that calls Jev for typed decisions.
Source: TypeSafe documentation
2026-09-20
Official models documentation lists jev-1.13.0 as the current version. Aliases jev-latest and jev-preview both pointed at that version when last verified. Input pricing was listed at $0.042 per million tokens, with output tokens free.
Source: TypeSafe Models
2026-09-18
Breaking change in typesafe-sdk 0.7.0: serialization moved from msgspec to pydantic, and system_one accepts an optional response_model. 0.7.1 (2026-09-21) validates the API key before the request and keeps it out of logged exceptions, and adds AI-gateway examples. Score.criteria has been an ordered list since 0.6.0 on both SDKs.
Source: TypeSafe Python SDK changelog
2026-09-17
TypeSafe documented known failure modes for jev-1.13, including literal reading, weak numeric precision, date comparison, context rot, and generation limits.
Source: TypeSafe documentation
2026-09-15
Third-party reporting described TypeSafe as a San Francisco lab leaving stealth with seed funding and Jev as its first System One model. Treat company-size and funding figures as third-party reports, not independently verified facts.
Source: Flavio Copes