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!

I can independently reproduce this issue on another Vercel Hobby team.

typesafe-ai/jev was working normally for me through AI Gateway, then began returning persistent HTTP 403 responses on September 27 without any intentional configuration change.

What I have verified:

• GET /v1/credits with the same AI Gateway key returns HTTP 200
• Current Gateway balance is approximately $4.998
• The API key therefore authenticates successfully and the team has available credit
• typesafe-ai/jev via the Evaluation API returns HTTP 403
• The failures occur across provider routing to both TypeSafe AI and DigitalOcean
• Before the failure, the same team/key/model combination was returning HTTP 200
• I have not changed the API key, team, model ID, or local integration in between the successful and failed requests

The transition in my logs is approximately:

• Sep 27 04:59 CEST: typesafe-ai/jev → HTTP 200
• Sep 27 06:59 CEST onward: typesafe-ai/jev → HTTP 403
• Persistent 403 responses since then

This appears consistent with the other reports here where Gateway authentication and credits work but Jev access specifically returns ACCESS_DENIED.

Could Vercel confirm whether there was an entitlement, plan-policy, provider-access, or provisioning change affecting typesafe-ai/jev around September 27?

I can provide specific gen_... request IDs from the failed requests if useful.

Got this response from vercel help :)

The AI Gateway team has suspended your team's AI Gateway access due to suspected abuse.

This block cannot be removed, and our support team is unable to review, override, or expedite this suspension.

I did ~20 requests of testing stuff. Love it...