Cloudflare gives Jev, TypeSafe’s structured evaluation model, its own page inside Cloudflare’s AI model catalogue, reachable at the model id typesafe/jev. You send it a state and a set of typed questions, and it returns calibrated answers with probabilities and confidence. The page lists the price: $0.042 per million input tokens, $0.00 for output and $0.00 for cached input, checked 20 October 2026.
What the integration does
This is a hosted copy of Jev sitting inside Cloudflare’s own catalogue of AI models, not a Cloudflare-built wrapper around TypeSafe’s API. Cloudflare’s page for typesafe/jev calls it “TypeSafe’s structured evaluation model” and describes it evaluating one state against typed Noul, Choice, and Score questions.
That is the same job Jev does everywhere else: read one piece of context, called the state, and answer a fixed set of questions about it with a number attached to each answer. Under the hood it works as a classification model, a kind of AI model that scores fixed options rather than generating open-ended text. Routing calls through Cloudflare keeps them on the same platform as other workloads there, much like the Vercel integration and the OpenRouter integration keep Jev inside their own stacks.
Official support
Cloudflare lists Jev in its AI model catalogue, at developers.cloudflare.com/ai/models/typesafe/jev/, which names the model id, the context window, and the request and response fields.
Jev does not show up in the general Workers AI models index at /workers-ai/models/. As of this writing, 20 September 2026, that index lists 65 models and does not mention Jev or TypeSafe anywhere. Jev is available on Cloudflare, just not on the index page most developers browse first, so searching there turns up nothing even though the dedicated page works.
Forbes reported that Cloudflare added Jev quickly after launch, in a piece by Josipa Majic. The cost and speed figures in that piece were not independently verified here, so treat it as a sign of how fast Cloudflare moved, not a confirmed number.
Clef on Workers AI
Cloudflare also trains its own decision models. Clef (27B) and Clef-flash (9B) launched on 1 October 2026 as @cf/cloudflare/clef and @cf/cloudflare/clef-flash, with model pages under /workers-ai/models/. They take the same state and typed questions as Jev, plus a required model field set to clef or clef-flash.
Call them from a Worker with env.AI.run("@cf/cloudflare/clef", {...}), or POST to https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/run/@cf/cloudflare/clef with a bearer token. A request takes 1 to 64 questions and, for both models, up to four images. The context window is 65,536 tokens for Clef and 24,576 for Clef-flash. Unlike the Jev listing, the price is published: $0.24 per million input tokens for Clef and $0.038 for Clef-flash, with no charge for output, checked on 10 October 2026. Clef-flash cost $0.09 at launch. Cloudflare says it does not read, store or train on requests or responses unless you opt into fine-tuning. The weights are also on Hugging Face under Apache 2.0.
A third model, Clef-omni, launched on 9 October 2026 as @cf/cloudflare/clef-omni. It is a mixture-of-experts model that takes optional audio and videos arrays as well as images, with a 64,000-token context, at $0.15 per million input tokens.
Cloudflare’s benchmark tables put Clef ahead of Jev on most tasks and behind on a few. Those numbers are Cloudflare’s own. The independent Jev Decision Index (multimodalart’s community board on Hugging Face; 0.4.1 file, generated 22 October 2026) puts Clef ahead of Jev on its public benchmarks and behind it on its Full score, which counts only private tests. Jev vs Clef sets out both.
Clairic on Workers AI
Clairic launched in preview on 19 October 2026 as @cf/cloudflare/clairic. It is not a decision model. Cloudflare calls it the first of a new class of System One model, extraction models. Instead of questions it takes named, typed fields, such as a text span, a number, a date, an estimate or a forecast, and returns a value for each, or null when the input does not determine it.
Call it the same way as Clef, with env.AI.run("@cf/cloudflare/clairic", {...}) or a POST to the matching /ai/run/ URL. It reads text and up to four images, has a 65,536-token context and costs $0.10 per million input tokens, with no output tokens, checked on 19 October 2026. Cloudflare says it intends to release the weights openly but had given no date as of 19 October 2026. Its own demo chains the two classes: Clairic reads the values, Clef decides on them and an LLM writes the reply. Jev vs Clairic and Clef vs Clairic cover how they fit together.
Setup
- Find the right page:
/ai/models/typesafe/jev/, not/workers-ai/models/. The Jev model page here covers the model itself. Cloudflare’s page covers only the hosted copy. - Use the model id
typesafe/jevwhen building the request. - Build the request body. Two fields are required:
state, the text or object being evaluated, andquestions, an array of typed questions. Each question needs atypeofnoul,choice, orscore, plusinstructionsandcriteria. - Mind the context window. Cloudflare’s listing gives Jev a 32,000-token context window.
- Read the response. Cloudflare returns
model,answers(shaped differently by question type), andusage, which reports input and output token counts. - Check pricing before committing to volume. Cloudflare’s page lists $0.042 per million input tokens and $0.00 for output, checked 20 October 2026. The speed and pricing guide covers TypeSafe’s own published figures.
- If you don’t have access yet, start with how to get Jev access before wiring up this request.
What people built with Jev and Cloudflare
Developers have used this route to run Jev checks straight from a Cloudflare Worker, and a few of those projects are linked below.
Limits and gotchas
The context window on this listing is capped at 32,000 tokens, per Cloudflare’s page, and the model is missing from the general Workers AI index, so searching there for Jev or TypeSafe won’t surface it even though the dedicated page works.
The listed price, $0.042 per million input tokens with no output charge, matches TypeSafe’s own rate. Check the page again before committing to volume.
The shape of answers depends on which question type you sent, so a batch mixing noul, choice, and score questions comes back with mixed answer shapes. Don’t assume one schema for the whole array.
FAQ
Can I use Clef on Cloudflare instead of Jev?
Yes for the request shape. Cloudflare says Clef is fully Jev-API compatible, so a state and typed Noul, Choice and Score questions run on both. Change the model id and add a model field, then rerun your own evals. Cloudflare’s tables put Clef ahead of Jev on most tasks and behind on a few.
Does Cloudflare support Jev officially?
Yes, in that Cloudflare itself hosts and documents the model on a dedicated page in its AI model catalogue, at /ai/models/typesafe/jev/. That page names the model id, the context window, and the request and response fields. It is not yet listed in Cloudflare’s general Workers AI models index, which as of 20 September 2026 lists 65 models without mentioning Jev or TypeSafe by name.
Do I need a TypeSafe API key to use Jev on Cloudflare?
Not confirmed either way. Cloudflare’s page for typesafe/jev documents the request and response shape and lists a price per million tokens, which suggests billing runs through Cloudflare rather than a separate TypeSafe account. Whether a TypeSafe key is also required isn’t stated on the page, so check the Jev model page and how to get Jev access before assuming either way.
What does Jev decide inside a Cloudflare Worker?
Whatever you ask it to, one state at a time. You send a state, the text or object you want evaluated, and a list of questions, each typed noul, choice, or score. Jev, described on Cloudflare’s page as TypeSafe’s structured evaluation model, returns a calibrated answer and a probability for each question you sent.
Is it free to use Jev on Cloudflare?
No. Cloudflare’s page for typesafe/jev lists $0.042 per million input tokens, with output and cached input at $0.00, checked 20 October 2026. No free tier is mentioned. TypeSafe’s own rate is in the speed and pricing guide if you are deciding between hosting routes.
Examples for Cloudflare
Introducing Clef: our open-source decision models, and new RL fine-tuning platform
Cloudflare's launch post for Clef and Clef-flash, two decision models on Workers AI that accept the Jev request format and open weights under Apache 2.0. It covers the architecture, Cloudflare's own benchmark tables against Jev, Kev and Laya, and a new reinforcement learning service.
Michelle Chen (@michellechen)140k viewsOpenIntroducing Clef: our open-source decision models, and new RL fine-tuning platform on blog.cloudflare.comcolinmcdermott/emoji-jev
Emoji autocomplete that makes one Jev call per pause in typing: a Choice over up to 254 emojis, a Choice over eight emotions, three Scores and three booleans, answered in parallel in about 100 ms of model time per the README. TanStack Start on Cloudflare Workers through the AI SDK typesafe-ai provider, hosted on Whop, with a live demo at emoji-jev.whop.site.
Introducing Clairic, a new System One class of extraction models
Cloudflare's launch post for Clairic, a preview extraction model on Workers AI that returns typed values such as spans, numbers, dates, estimates and forecasts with probabilities. It sets out extraction as a second System One class beside decision models and gives Cloudflare's own forecast and claims-pipeline results.
Cloudflare122k viewsOpenIntroducing Clairic, a new System One class of extraction models on blog.cloudflare.com
Introducing Clef-omni with full multimodality, plus a faster Clef and a cheaper Clef-flash
Cloudflare's launch post for Clef-omni, a 30B-A3B mixture-of-experts decision model that reads audio, video, images and text, at $0.15 per million input tokens. It also cuts Clef-flash to $0.038 with a 24K hosted context, and gives Cloudflare's own benchmark and latency tables.
Jev on Cloudflare Workers AI
Cloudflare Workers AI model catalog page for TypeSafe Jev, documenting the state plus questions request shape and the Noul, Choice and Score answer types. Examples cover ticket routing, refund review and risk assessment.