--- title: TypeSafe AI Releases Jev: A Decision-Only Model That... description: TypeSafe AI, led by former OpenAI researcher Diogo Almeida, has launched Jev, a decision-only model that returns typed probabilistic outputs (Choice, Score,... canonical: https://daily.dev/posts/typesafe-ai-releases-jev-a-decision-only-model-that-returns-typed-probabilities-instead-of-text-zci7cgwtw twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: TypeSafe AI Releases Jev: A Decision-Only Model That Returns Typed Probabilities Instead of Text | daily.dev og:description: TypeSafe AI, led by former OpenAI researcher Diogo Almeida, has launched Jev, a decision-only model that returns typed probabilistic outputs (Choice, Score,... og:url: https://daily.dev/posts/typesafe-ai-releases-jev-a-decision-only-model-that-returns-typed-probabilities-instead-of-text-zci7cgwtw og:image: https://api.daily.dev/og/posts/zCi7cGwTw.png og:image:alt: TypeSafe AI Releases Jev: A Decision-Only Model That Returns Typed Probabilities Instead of Text og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # TypeSafe AI Releases Jev: A Decision-Only Model That Returns Typed Probabilities Instead of Text **[InfoQ](https://daily.dev/sources/infoq)** · 4 min read · 0 upvotes · 0 comments ## Summary TypeSafe AI, led by former OpenAI researcher Diogo Almeida, has launched Jev, a decision-only model that returns typed probabilistic outputs (Choice, Score, Noul) instead of generated text, aimed at classification, routing, and scoring tasks. It evaluates typed questions in a single parallel pass with a 32K token context window, $0.042 per million input tokens, free output, and 70-500ms latency. Vercel added it to AI Gateway on day two and reported adoption by nearly 13% of paid teams within 24 hours, twice the rate of the GPT-5.6 family. Netlify and LangChain integrated it as well, and developers report speedups of 5-18x over comparable LLM classifiers, though independent analysis found median real-world speedups closer to 7x against a 193.6x headline figure. Critics note Jev can't emit invalid types but can still return confidently wrong valid values, and its documented 'jaggedness' includes unreliable counting, arithmetic, and date comparisons. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is Jev from TypeSafe AI and how is it different from a regular LLM? Jev is a decision-only model from TypeSafe AI that returns typed, probabilistic outputs like Choice, Score, and Noul instead of generated text, so software can act directly on a probability and confidence value rather than parsing free text. It evaluates a state plus typed questions in one parallel pass, costs $0.042 per million input tokens with free output, has a 32,000 token context window, and quotes 70-500ms end-to-end latency. _Developers evaluating classification and routing models can track new releases like this one on daily.dev._ ### How much faster is Jev than an LLM for classification tasks like safety filtering or email sorting? A Vercel engineer found a safety classifier running on Jev performed five to 18 times faster than the LLM it replaced, and a CTO comparing it to Gemini for email classification found Gemini slightly more accurate but 10 to 20 times more expensive. An independent analysis of over 12,000 launch tweets found user-reported speedups at a median of 7x, far below the 193.6x headline figure, with median cost savings of 30x. _Teams weighing model speed versus accuracy tradeoffs can compare real-world benchmarks like these on daily.dev._ ### Can Jev hallucinate like a traditional LLM? Jev cannot emit an invalid type, since its outputs are constrained to typed structures like Choice and Score, but it can still return a completely wrong value that is technically valid for that type. TypeSafe documents known jaggedness in jev-1.13, including unreliable counting, arithmetic, and date comparison, and advises keeping math logic in code rather than relying on the model. _Developers building reliability checks around new AI models can dig into tradeoffs like this on daily.dev._ --- Tags: [#vercel](https://daily.dev/tags/vercel), [#ai-gateway](https://daily.dev/tags/ai-gateway) [View this post on daily.dev](https://daily.dev/posts/typesafe-ai-releases-jev-a-decision-only-model-that-returns-typed-probabilities-instead-of-text-zci7cgwtw) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"TypeSafe AI Releases Jev: A Decision-Only Model That Returns Typed Probabilities Instead of Text","url":"https://daily.dev/posts/typesafe-ai-releases-jev-a-decision-only-model-that-returns-typed-probabilities-instead-of-text-zci7cgwtw","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/typesafe-ai-releases-jev-a-decision-only-model-that-returns-typed-probabilities-instead-of-text-zci7cgwtw"},"datePublished":"2026-10-01T07:07:34.186Z","dateModified":"2026-10-02T05:00:19.141Z","description":"TypeSafe AI, led by former OpenAI researcher Diogo Almeida, has launched Jev, a decision-only model that returns typed probabilistic outputs (Choice, Score,...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/62ecefdca2d9607239918efa75bd3ddb?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/62ecefdca2d9607239918efa75bd3ddb?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"InfoQ","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"InfoQ","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/afc3bced3e1e4b188dd9127017a60e0c","url":"https://daily.dev/sources/infoq"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/typesafe-ai-releases-jev-a-decision-only-model-that-returns-typed-probabilities-instead-of-text-zci7cgwtw","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"vercel,ai-gateway","timeRequired":"PT4M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"InfoQ","item":"https://daily.dev/sources/infoq"},{"@type":"ListItem","position":3,"name":"TypeSafe AI Releases Jev: A Decision-Only Model That Returns Typed Probabilities Instead of Text"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/typesafe-ai-releases-jev-a-decision-only-model-that-returns-typed-probabilities-instead-of-text-zci7cgwtw#faq","mainEntity":[{"@type":"Question","name":"What is Jev from TypeSafe AI and how is it different from a regular LLM?","acceptedAnswer":{"@type":"Answer","text":"Jev is a decision-only model from TypeSafe AI that returns typed, probabilistic outputs like Choice, Score, and Noul instead of generated text, so software can act directly on a probability and confidence value rather than parsing free text. It evaluates a state plus typed questions in one parallel pass, costs $0.042 per million input tokens with free output, has a 32,000 token context window, and quotes 70-500ms end-to-end latency. Developers evaluating classification and routing models can track new releases like this one on daily.dev."}},{"@type":"Question","name":"How much faster is Jev than an LLM for classification tasks like safety filtering or email sorting?","acceptedAnswer":{"@type":"Answer","text":"A Vercel engineer found a safety classifier running on Jev performed five to 18 times faster than the LLM it replaced, and a CTO comparing it to Gemini for email classification found Gemini slightly more accurate but 10 to 20 times more expensive. An independent analysis of over 12,000 launch tweets found user-reported speedups at a median of 7x, far below the 193.6x headline figure, with median cost savings of 30x. Teams weighing model speed versus accuracy tradeoffs can compare real-world benchmarks like these on daily.dev."}},{"@type":"Question","name":"Can Jev hallucinate like a traditional LLM?","acceptedAnswer":{"@type":"Answer","text":"Jev cannot emit an invalid type, since its outputs are constrained to typed structures like Choice and Score, but it can still return a completely wrong value that is technically valid for that type. TypeSafe documents known jaggedness in jev-1.13, including unreliable counting, arithmetic, and date comparison, and advises keeping math logic in code rather than relying on the model. Developers building reliability checks around new AI models can dig into tradeoffs like this on daily.dev."}}]} ```