Agents: read the raw Markdown of this page, or start at llms.txt.
Blog: Jev 1.14 (2026-10-14)
TL;DR TypeSafe's 2026-10-14 post releases
jev-1.14.0underjev-preview. It describes 1.14 as small improvements, with the largest gain expected in multilingual support. It also adds three experimental features behind "danger zone" request headers: images (on 1.13 too), theboxquestion type (1.14 only) and long context. TypeSafe says these are not ready for production.jev-latestmoves tojev-1.14.0"in a week" (about 2026-10-21). TypeSafe intends 60 days' notice before deprecatingjev-1.13.0.
Facts
| Field | Value |
|---|---|
| Title | "Jev: For the builders AND dreamers" |
| URL | https://typesafe.ai/blog/jev-1-14 |
| Category | Company News |
| Date | 2026-10-14 (page shows "Oct 14, 2026") |
| Author | Not named on the page; written as "we" |
| Model released | jev-1.14.0, under the jev-preview alias |
| Alias move | "jev-latest will be updated to jev-1.14.0 in a week." "jev-1.13.0 will be available for a while." |
| Deprecation intent | 60 days' warning before jev-1.13.0 is formally deprecated, 30 days for jev-1.14.0; "We'll release a deprecation policy prior to this" (no policy in the docs as of 2026-10-15) |
| Captured | 2026-10-15 (text and HTML) |
What it announces
Danger-zone flags. The post sets out a split between "prod builders" and "dreamers". Features that TypeSafe thinks are good but too jagged to vouch for go behind TypeSafe-Danger-Zone-* request headers. TypeSafe says these features "will be here to stay for future models". It also says "there is still way too much jaggedness for us to vouch for these for production use cases".
| Feature | Header | Scope (from the post and the docs) |
|---|---|---|
| Images | TypeSafe-Danger-Zone-Images: true |
jev-1.13.0 and jev-1.14.0. Images go in a top-level assets map and are referenced by key from state, instructions or criteria. The post says 1.13 "was multimodal all along". It warns that all models "perform worse on multimodality over pure text" and that splitting questions into the simplest parts helps. Details: raw/docs/images.md |
box question type |
Same images header | jev-1.14.0 only. It returns one bounding box, the post's "first type that's not purely meant for smart-if statements". Details: raw/docs/primitives__box.md |
| Long context | TypeSafe-Danger-Zone-Long-Context: true |
The post says Jev "supports a much longer context length than we've enabled in production". The docs give 262k tokens for both limits, at the same price, and call it "slower and less accurate" (Models, aliases, pricing, rate limits, context) |
The model. jev-1.14.0 is "Not meant to be flashy but instead be small, subtle improvements to everything". "Most things should be a tiny bit better, but we expect the largest gain in multi-lingual support." These are TypeSafe's claims. Neither the post nor the docs publish per-language numbers. The models page gives 1.14 the same price and default context limits as 1.13.
Opt-in query sharing. TypeSafe added "an opt-in to allow you to help find model jaggedness with your queries". It adds: "We will still not train on your queries". The post does not say where the setting lives. The image placed after this paragraph has no alt text, so its content is not captured.
Its claims, attributed
- Internal intelligence chart. The "Measuring Intelligence" section shares TypeSafe's "own internal best guesses" at measuring intelligence for developers. The post attaches its own caveats: the measurement "will continue to evolve" and future numbers may not be consistent; "Don't believe it entirely"; only open-weight models are included because TypeSafe guards its validation set and worries about providers' data policies; and "This is likely biased towards our models. Feel free to ignore our model, and instead focus on the relative ranking of open source models." The two chart images have no alt text in the captured HTML, so this wiki does not record which models or numbers they show.
- The post's 2026-09-11 predecessor promised "no standard benchmark table in our model releases" (Blog: Lies, Damned Lies, and Benchmarks). This chart is labelled an internal measurement, not a public benchmark. The post links to that predecessor ("we dislike public benchmarks").
- What's next. TypeSafe expects most releases to be "beautifully boring": "small improvements across the board, better stability, and features for real deployment".
Relationship to Jev
This post is the primary source for the 1.14 release, the alias-move timing and the deprecation intent. The docs carry the contract details: models.md for versions, aliases and long context; images.md and primitives/box.md for the experimental features. As of 2026-10-15 the docs disagree with each other in places. The models table still lists Input "Text" for both versions, while images are accepted behind the header. See Jev (model) and Versions and timeline (models, SDKs, API, company).
Related
- Jev (model) — model facts, aliases and version history
- Models, aliases, pricing, rate limits, context — price, context limits, long context, aliases
- Running Jev in production: versioning, caching, retries, monitoring and fallbacks — pinning versions across the alias move
- Blog: Lies, Damned Lies, and Benchmarks — TypeSafe's earlier stance on public benchmarks
- Versions and timeline (models, SDKs, API, company) — dated record of releases
Sources
- raw/site/blog-jev-1-14.txt (https://typesafe.ai/blog/jev-1-14), captured 2026-10-15
- raw/site/blog-jev-1-14.html: model IDs in the alias and deprecation sentences, which the text extract drops; the five images have no alt text
- raw/docs/models.md (https://docs.typesafe.ai/models)
- raw/docs/images.md (https://docs.typesafe.ai/images)
- raw/docs/primitives__box.md (https://docs.typesafe.ai/primitives/box)