# Netlify's AI Gateway Gets a Model Built to Decide, Not Chat > Source: > Published: 2026-09-21 00:00:00+00:00 # Netlify's AI Gateway Gets a Model Built to Decide, Not Chat Netlify's AI Gateway added TypeSafe's Jev, a model returning typed choices and probabilities instead of text, a day after Vercel shipped it too. Netlify’s AI Gateway can now reach a model that never writes a sentence, only typed decisions. TypeSafe’s Jev went live there at zero setup, Netlify said in a Sept. 17 changelog entry, the gateway’s first model built for classification and routing rather than conversation ([Netlify changelog, “TypeSafe Jev now available in AI Gateway,” Sept. 17, 2026](https://www.netlify.com/changelog/typesafe-jev-ai-gateway/)). ## What changed Jev is TypeSafe’s first “System One” model, and it answers only three kinds of question: `choice` picks one option from a declared set, `score` rates against ordered levels, and `noul` returns a yes/no probability between 0 and 1. The answer shape is fixed in advance, so there’s no JSON to parse or schema to coerce on the way out. The model evaluates multiple questions in parallel against roughly 150,000 characters of shared context, with response times of 70 to 500 milliseconds, Netlify’s changelog said. Running it means installing `@typesafe-ai/sdk` inside a Netlify Function; there’s no API key to create, and usage bills against the same Netlify credits as any other AI Gateway model. It needs Node.js 20 or newer and defaults to the `jev-latest` alias, currently `jev-1.13.0`. Netlify wasn’t first. Vercel added Jev to AI SDK 7 a day earlier, through an experimental `evaluate()` API exposing the same three typed primitives plus a `zeroDataRetention` gateway option (Vercel changelog, “TypeSafe AI’s Jev now available on AI Gateway,” Sept. 16, 2026). TypeSafe frames Jev’s use cases as routing decisions inside agent loops, choosing whether to escalate a task, and validating another model’s output: plumbing work, not chat. Both integrations support Zero Data Retention and no-training guarantees per request, a detail that matters more for compliance-sensitive agent pipelines than for a general chatbot. ## Why it matters Every model Netlify has added to AI Gateway this year, including Claude Fable 5.1, Gemini 3.8 Flash and DeepSeek V4.1 Flash, has been a conversational model reachable through a chat-completions-shaped API (Firerun, [Sept. 8, 2026](https://firerun.io/netlify-ai-gateway-claude-gemini-2026/)). Jev breaks that pattern: it’s the first gateway addition that doesn’t generate text at all. That it landed on both Netlify and Vercel within a day of each other, through each platform’s own integration rather than a shared wrapper, suggests both are treating small typed-decision models as a distinct category worth building for, not a novelty add-on. TypeSafe’s own numbers should be read as a vendor claim, not an independent finding: the company says Jev was “up to 193.6x faster and 444.6x cheaper than LLMs on its workflow evaluations,” a comparison run on TypeSafe’s own benchmark suite against unspecified baseline models. Netlify and Vercel are relaying that figure, not verifying it. ## The take A model that only picks from an answer set sounds like a step down from a chatbot, but for the job it’s built for — is this ticket urgent, does this output pass a guardrail, which queue should this request join. A bounded, typed answer beats free text a caller still has to parse. Landing on two competing gateways in the same week reads less like one vendor’s launch and more like both platforms betting agent infrastructure needs decision primitives, not just bigger chat models. Whether Jev holds that ground long-term is a separate question from whether the category does. ## Key Takeaways - TypeSafe’s Jev reached Netlify’s AI Gateway Sept. 17 with zero setup, and Vercel’s AI SDK 7 a day earlier, via `@typesafe-ai/sdk` and an experimental`evaluate()` API respectively. - Jev answers only three question types — `choice` ,`score` and`noul` (a yes/no probability) — evaluated in parallel with 70- to 500-millisecond response times, no generated text involved. - Both integrations support per-request Zero Data Retention and no-training guarantees, aimed at agent pipelines handling sensitive state. - TypeSafe’s claim of up to 193.6x speed and 444.6x cost gains over LLMs comes from its own benchmark suite and hasn’t been independently verified.