AI Gateway Jev request returns 403 and is absent from logs

Hi Vercel team,

I’m trying to use the AI Gateway Evaluation API with typesafe-ai/jev on the Hobby team griffin-detrick-6479.

What I tested:

Created a new AI Gateway API key named jev-local-playground.
Sent POST https://ai-gateway.vercel.sh/v1/evaluate with Bearer authentication, Content-Type: application/json, and model typesafe-ai/jev.
The request used a valid choice question with three criteria.
AI Gateway returned HTTP 403.
The rejected request does not appear in AI Gateway Logs, although the key’s dashboard entry shows recent use.
Creating a fresh key and retrying produced the same result.

No API key or secret is included in this post. The team has no prior Gateway usage. The published pricing documentation says every Vercel team account receives free-tier AI Gateway access.

Could someone check whether this team’s AI Gateway entitlement or provisioning is incomplete and enable the Evaluation API if appropriate?

Thanks.

Same issue here on a Hobby team. I reproduced it independently and isolated the failure to Jev / Evaluation API. Working on the same team and API key:

GET /v1/models → HTTP 200
GET /v1/credits → balance $5
POST /v1/chat/completions with openai/gpt-4.1-nano → HTTP 200 and billed successfully
typesafe-ai/jev is present in the model catalog Failing:
POST /v1/evaluate, model typesafe-ai/jev → HTTP 403
POST /typesafe/v1/systemone, model jev-latest → HTTP 403 I also verified:
Provider Allowlist is disabled
Zero Data Retention is disabled
vercel ai-gateway rules ls returns No routing rules found Both failures reproduce with direct PowerShell HTTP requests, without SDK/application code. This therefore appears to be a Jev / Evaluation API entitlement or provisioning issue on Hobby teams rather than an authentication, billing, routing-rule, or application problem. Could Vercel staff check the Jev/Evaluation entitlement for affected Hobby teams?

Hi, I'm experiencing the same issue and have carried out some additional troubleshooting that may help narrow it down.

My setup is Windows PowerShell with Vercel CLI 60.1.3.

I've confirmed the following:

Jev (typesafe-ai/jev) appears in the Gateway model catalogue.
My API key successfully authenticates against /v1/credits.
My account has $5 available credit and an active payment method.
A standard request using openai/gpt-4o-mini succeeds with the same API key.
Requests to /v1/evaluate using Jev consistently return HTTP 403, with no additional error details.
One subsequent retry returned HTTP 429.

Given that standard AI Gateway requests work correctly, and another user has reproduced the same Jev issue with a freshly generated API key, this appears potentially related to Jev-specific access, Evaluation API provisioning, or a service-side issue.

Could Vercel confirm whether this is a known issue or whether additional permissions are required for the Evaluation API?

Thanks!