Jev Channels Compared: TypeSafe, Vercel, OpenRouter, Cloudflare
Last checked · Independent guide, not affiliated with TypeSafe AI
Jev is available on TypeSafe's own API and, as of September 19, 2026, through Vercel AI Gateway, OpenRouter and Cloudflare Workers AI. They serve the same Jev 1.13 model but differ in model name, request format, context window, price and what you need before the first call.
Within four days of launch, Jev could be called in four places. They all run Jev 1.13, but you cannot swap one for another by changing a URL: each has its own model name, request shape and limits.
Side-by-side
Section titled “Side-by-side”| Feature | TypeSafe API | Vercel AI Gateway | OpenRouter | Cloudflare Workers AI |
|---|---|---|---|---|
| Model name | jev-1.13.0, jev-latest, jev-preview |
typesafe-ai/jev |
typesafe/jev-1.13 |
typesafe/jev |
| How you call it | POST https://api.typesafe.ai/v1/systemone or the official SDKs |
AI SDK 7+ experimental_evaluate |
POST https://openrouter.ai/api/alpha/decisions |
env.AI.run() in a Worker, or the Workers AI REST API |
| Yes/no question type | noul, answer field noul |
boolean, answer field probability |
noul |
noul |
| Context window | 64k per request; 32k for state + longest question | Not stated on the model page | 32,000 tokens | 32,000 tokens |
| Price (input / output) | $0.042 per 1M / free | Listed free until Sep 25, 2026 (promotion) | $0.042 per 1M / $0 | In the Cloudflare dashboard |
| Before your first call | Account approved from the waitlist | Card on file for the Vercel team | Prepaid credits | Cloudflare account and API token |
| Listed since | Sep 15, 2026 | Sep 15, 2026 | Sep 18, 2026 | Sep 19, 2026 |
| Tested by us | Yes, Sep 19 | Yes, Sep 18 | Endpoint checked, not called | Not yet |
Which one should you use?
Section titled “Which one should you use?”- To try Jev in the next ten minutes: OpenRouter. Top up a few dollars and send a request; the official request format is almost identical.
- To build something you will keep: the TypeSafe API. It has the biggest context window, the official Python and JavaScript SDKs, model version pinning and TypeSafe’s status page. You may have to wait for access.
- If your app already runs on Vercel and the AI SDK: Vercel AI Gateway, as long as you add a card. Watch the free-tier rate limits and the end of the promotion on September 25.
- If you run on Cloudflare Workers: Workers AI keeps the call inside Cloudflare’s network. Check the price in your dashboard first.
The differences that break code
Section titled “The differences that break code”1. The yes/no type has two names. Vercel’s AI SDK calls it boolean and returns probability. TypeSafe’s API calls it noul and returns noul. Sending "type": "boolean" to TypeSafe’s endpoint returns HTTP 400 “Invalid request.” See 400 Invalid request.
2. Context is halved on two channels. OpenRouter and Cloudflare both list a 32,000-token context window. TypeSafe documents 64k per request, with the same 32k cap on the state plus the longest question. In our tests, TypeSafe rejected a state of about 33,600 tokens with max_tokens_exceeded, so in practice the usable state size is similar everywhere.
3. Model names are not portable. jev-1.13 (without the patch number) is rejected by TypeSafe with “Unknown model”, while typesafe/jev-1.13 is the correct name on OpenRouter. OpenRouter’s API reference shows responses reporting a dated version, such as typesafe/jev-1.13-20260917.
4. Errors look different. TypeSafe returns {"detail": ...}; OpenRouter returns {"error": {"code": ..., "message": ...}}; Vercel’s SDK throws gateway errors. The errors index lists each one with its channel.
Each channel in detail
Section titled “Each channel in detail”- Jev API (TypeSafe): request format, responses, errors
- Jev on OpenRouter
- Jev on Vercel AI Gateway
- Jev on Cloudflare Workers AI
- How to get a Jev API key: the sign-up side of each route