Missing x-ratelimit-remaining-requests in response for embeddings generation

Hi Roman Egel

The absence of the x-ratelimit-remaining-requests header in the response for embeddings generation could be due to changes in the rate limiting implementation. As of January 2025, rate limits are based on Requests-Per-Minute (RPM) and Tokens-Per-Minute (TPM) for the gpt-4o-realtime-preview model. That may have shifted to reporting only the remaining tokens instead of both requests and tokens.

If you are only seeing x-ratelimit-remaining-tokens, it may indicate that the service is currently prioritizing token-based rate limiting, and the requests-based limit may not be reported in the same way as before.

you can keep an eye on the Azure OpenAI Service release notes for updates on What's new in Azure OpenAI Service page will be updated accordingly with more details.

Kindly refer below link: understanding-rate-limits

Thank You.