Native audio input support for Gemini and Muse models #1131

Feature Description

Is native audio input support planned for Gemini and Muse models?

Sending WAV audio as input_audio to Command Code’s Provider API (POST /provider/v1/chat/completions) returns HTTP 400 (Invalid input, messages.0.content) with both google/gemini-4-argon and meta/muse-ember.

Could Command Code support native audio input or provide a compatible endpoint for these models? Is there an expected timeline?

Use Case

Direct audio transcription, translation, and timestamped subtitle generation using the model’s audio understanding.

Additional Context

For comparison, Gemini 4 Argon successfully transcribes audio through OpenCode’s Google-native endpoint using inlineData, with AUDIO input tokens confirmed in the response:

POST https://opencode.ai/inference/go/google/v1beta/models/gemini-4-argon:generateContent

The audio is passed as Base64 WAV data in contents[].parts[].inlineData with mimeType: "audio/wav".

Related: #906

How important is this to you?

Important for my workflow