Jev on Vercel AI Gateway: Setup, Pricing and Limits

Vercel listed Jev in its AI Gateway on launch day, September 15, 2026. We used this route for our first 131-question test of Jev before TypeSafe’s own API opened for our account, so this page is based on real runs as well as Vercel’s documentation.

At a glance

Item	Value
Model	typesafe-ai/jev
How to call	AI SDK (ai package) version 7 or later, experimental_evaluate; or the TypeSafe-compatible API at https://ai-gateway.vercel.sh/typesafe
Auth	AI Gateway API key (AI_GATEWAY_API_KEY)
Price	$0.042 per million input tokens, output free (the free launch promotion ended Sep 25, 2026)
Context window	32,000 tokens
Providers behind it	TypeSafe AI; DigitalOcean added Sep 24, 2026
Account requirement	A payment card on the Vercel team, even for free-tier use
Yes/no question type	boolean, which returns probability
OpenAI-compatible endpoint	Not available for Jev
Bring your own key	Yes, add a TypeSafe key under BYOK to be billed by TypeSafe directly

Step 1: Add a card before you do anything else

This is where most first attempts fail. A new Vercel team with no payment method gets an error containing customer_verification_required when it calls Jev. Adding a card to the team unlocks the $5 of free monthly credit, which covers a lot of Jev requests at $0.042 per million input tokens. You do not have to buy credits.

Be careful with the next step: buying credits changes your plan. Vercel’s pricing docs say that purchasing AI Gateway credits moves the team to the paid tier, which raises rate limits but ends the $5 monthly free credit. If you only want to experiment, add a card and do not top up. In the dashboard, the credit balance sits behind the balance button at the top right of the AI Gateway page.

Step 2: Make the call

Install the AI SDK and set your key:

npm install ai
export AI_GATEWAY_API_KEY="your-gateway-key"

Then call Jev with experimental_evaluate. This is the pattern Vercel shows on its model page, adapted to our support-ticket example:

evaluate.mjs

import { experimental_evaluate as evaluate } from 'ai';

const result = await evaluate({
  model: 'typesafe-ai/jev',
  state: 'I was charged twice for my annual plan this morning. Please refund one of the charges today.',
  questions: {
    wantsRefund: {
      type: 'boolean',
      instructions: 'Is the customer asking for money back?',
    },
  },
});

console.log(result.answers.wantsRefund); // a boolean answer with a probability field

Vercel describes Jev as returning choices, scores and boolean probabilities. The yes/no type is the one we know is named differently: boolean here, noul on TypeSafe. If you later move to TypeSafe’s endpoint and keep "type": "boolean", you will get HTTP 400 “Invalid request.” (see 400 Invalid request).

Or: keep the official TypeSafe SDK

Since September 21, 2026, Vercel also serves a TypeSafe-compatible API. Point the official SDK at it by changing the base URL and key, and use the model name typesafe-ai/jev:

typesafe-sdk-via-vercel.mjs

import { TypeSafeClient } from '@typesafe-ai/sdk';

const client = new TypeSafeClient({
  apiKey: process.env.AI_GATEWAY_API_KEY,
  baseURL: 'https://ai-gateway.vercel.sh/typesafe',
});

const result = await client.systemOne({
  model: 'typesafe-ai/jev',
  state: 'I was charged twice for my annual plan this morning.',
  questions: {
    refund: { type: 'noul', instructions: 'Is the customer asking for money back?' },
  },
});

This route uses TypeSafe’s names, so the yes/no type is noul, not boolean. It supports POST /typesafe/v1/systemone and GET /typesafe/v1/models. The same trick works from Python: TypeSafe’s Python SDK documentation shows base_url="https://ai-gateway.vercel.sh/typesafe" with model="typesafe-ai/jev". Requests are billed through AI Gateway unless you add your own TypeSafe key under BYOK. Vercel recommends its own evaluation API for new code; the TypeSafe-compatible one is meant for moving existing code.

The free tier is slow for batch work

Vercel’s docs say free-tier requests are rate-limited per model at lower limits than the paid tier, and that exceeding them returns a 429. For Jev, we found the limit very tight: a few questions would go through, then requests were throttled for a while. Our 131-question batch took about an hour and a half on the free tier, while 145 similar questions took 81 seconds on TypeSafe’s own API.

In code, the AI SDK surfaces this as a gateway rate-limit error; the HTTP status is not always exposed on the error object, so match on the error type or message too. See 429 Too many requests for a backoff pattern.

Pricing after the launch promotion

Vercel listed Jev as free for input and output from launch until September 25, 2026. After the promotion, Vercel’s model list prices Jev at $0.042 per million input tokens with free output, the same as TypeSafe and OpenRouter. Vercel does not add a markup, but buying AI Gateway credits ends the free-tier credit as described above. See Is Jev free? for the free options that are left.

On September 24, Vercel added DigitalOcean as a second provider for typesafe-ai/jev. Vercel marks TypeSafe’s own endpoint as supporting zero data retention and DigitalOcean’s as not, so if retention matters, check the provider routing options in your gateway settings.

Pros and cons of this route

Good for: apps already on Vercel and the AI SDK, one bill for all model providers, and using Jev while TypeSafe’s own signups are paused.

Watch out for: the card requirement, the free-tier throttle, the different yes/no type name, and relying on an API that the AI SDK still marks experimental.

Compare with the other routes in Jev channels compared.