New in Evals: Full Audio Support

Evals now supports grading audio responses directly from audio inputs.

This enables evaluations in scenarios where both the user and the agent communicate through audio.

The cookbook example shows how to:

Provide audio files as input to your evals, skipping transcription entirely

Configure evals to handle model audio outputs

Leverage text transcripts alongside the existing suite of text graders

Feel free to ask questions, share your experiences, or get advice from the community as you explore this feature.

3 Likes