Jev: For the builders AND dreamers - TypeSafe AI Blog
Short summary
TypeSafe released Jev 1.14.0 with a split approach: stable improvements for production use and experimental "danger zone" features (image support, bounding box type, longer context) for developers willing to accept less reliability. The company is also sharing internal intelligence metrics while acknowledging their limitations.
- Image support is now composable across Jev's JSON interface (state, instructions, criteria), though multimodal performance is weaker than text-only and requires careful prompt decomposition.
- New "box" type returns bounding boxes without type errors, representing the first Jev type designed beyond binary decisions and signaling exploration into broader System One Model capabilities.
- Jev supports longer context lengths than previously enabled in production; developers can opt in via a danger zone flag, but TypeSafe intentionally limits context by default because reliability degrades at scale.
- Internal benchmarking data is being shared for the first time, limited to open-weight models only, with explicit caveats that the methodology will evolve and developers should run their own evals.
Full summary
TypeSafe released Jev 1.14.0 with a deliberate split: stable features for production use and experimental capabilities marked as "danger zone" for developers who want to experiment. The company's reasoning is straightforward—they can't predict what creative builders need, so they're shipping less-tested features alongside their battle-tested core. Three experimental features arrive with this version: image support (fully composable, works in state, instructions, or criteria), a new "box" type that returns bounding boxes for non-decision use cases, and optional longer context windows. All three come with honest caveats: images degrade model performance, long contexts hurt reliability, and the box type is still rough around the edges. You enable these with API headers like `TypeSafe-Danger-Zone-Images: true`.
The 1.14.0 release itself brings incremental improvements across the board, with multilingual support getting the biggest boost. TypeSafe is staggering the rollout—jev-1.14.0 becomes the default in a week, while 1.13.0 stays available for at least 60 days before deprecation. They're committing to 30 days' notice before removing 1.14.0. The company also added an opt-in telemetry system that lets users help surface model failures without training on user queries, and they're sharing internal intelligence measurements (though with heavy disclaimers that these are best guesses and shouldn't be trusted absolutely).
The messaging reflects TypeSafe's frustration with how developers hate updating dependencies. They're treating users like adults by being transparent about limitations and sharing internal metrics rather than hiding behind polished benchmarks. They explicitly say to run your own evals, don't believe their numbers entirely, and they're only comparing against open-weight models because they won't touch other companies' data policies. Future releases will apparently stay boring—small stability wins and deployment features rather than flashy announcements.