Jev 1.14 from @typesafeai is now available through AI Gateway. 𝚝𝚢𝚙𝚎𝚜𝚊𝚏𝚎-𝚊𝚒/𝚓𝚎𝚟-𝟷.𝟷𝟺.𝟶 • Ask typed questions about images • Enable experimental long context • Both features are supported in Jev 1.13

Jev images is here One question about an image that is 850x850 cost one hundredth of a cent - $0.000096936 about 10,000 images for a dollar. referencing the image more than once duplicates the token usage - so asking two questions in a single request is equivalent to uploading two images

Jev + Opus 5.5 = the BEST stack for a company brain. This 14-page PDF covers everything you need to know. The smart ones will send this to their agent: →

JEV-27B-VL brings accurate, calibrated visual decisions to long-context workflows. 🤖 📄 🏆 Ranks #1# among 20 models on the independent Decision Index Vision board, with 77.8% accuracy across 13,695 examples. 🎯 Its confidence scores closely track real accuracy, achieving an ECE of 0.019 for reliable threshold-based decisions. 📚 System 1 supports image-and-text prompts up to 256K tokens, with yes/no, 0–5 scoring, or selection from as many as 256 options. ⚡ Each System 1 query returns a calibrated probability for every option in one forward pass, while System 2 handles step-by-step multimodal reasoning. 👁️ The decision adapter was trained only on text, yet its zero-shot vision performance exceeds the 397B Qwen3.5 reference by 6.4 points on the board’s full score. 📜 Open weights under Apache 2.0.

Jev Skill Typeahead for Claude Code As you type your prompt, Jev analyzes it and predicts which of your installed skills Claude is most likely to invoke It shows each skill's estimated probability before you even send the request Super useful for understanding and optimizing your skill pipelines before execution And it uses 0 Claude tokens Install: npx claude-code-templates@latest --mod productivity/jev-skill-typeahead

Jev just got competition from Amazon and Cloudflare. Both have released open-source decision models. Like TypeSafe’s Jev, they pick from choices you supply and return probabilities. Give them a support ticket and ask which team should handle it, or an agent’s proposed action and ask whether it should proceed. No paragraph of generated text required. Amazon’s Strands Decider has just 2 billion parameters and runs locally on a CPU or GPU. Amazon released the weights, training data and training scripts, and reports roughly 153 ms median latency for small tasks on an M3 MacBook. Cloudflare’s Clef comes in 27B and 9B versions, supports images and doubles Jev’s context window to 64K. It also supports Jev’s API format. In Cloudflare’s own tests, Clef beat Jev on three of four workflow evaluations, while Clef-flash returned decisions in about 39 ms versus Jev’s 524 ms median.

Jev is now in the AI SDK for Python. To test it, we ran two experiments: ▪︎ Detecting Python vs. English as text is typed ▪︎ Writing Python, one decision at a time 𝚞𝚟 𝚊𝚍𝚍 𝚊𝚒