Guides › How to Configure the Whisper Model in OpenCode

How to Configure the Whisper Model in OpenCode

The default whisper base model is fast but can struggle with accents or technical terms. Free tier uses base only. Pro ($29 one-time) lets you switch to small, medium, or large for better accuracy.

Download a different model
Run MODEL=small bash scripts/install-whisper.sh to download the small model. Replace "small" with "medium" or "large" for larger models.

Set the model path
Set the WHISPER_MODEL env var: export WHISPER_MODEL="$HOME/.local/bin/ggml-small.bin". Add this to ~/.bashrc or ~/.zshrc for persistence.

Or use plugin config (Pro only)
In opencode.json: [["opencode-interrupt-plugin", { "licenseKey": "INTERRUPT-...", "whisperModel": "small" }]]. This accepts "base", "small", "medium", "large", or a full file path. Free users who set whisperModel in config will still get base.

Restart OpenCode
Restart opencode for the change to take effect. The plugin will use the new model for all future /ptt transcriptions.

Tips
• Model accuracy vs speed: base < small < medium < large. Small is the best balance for most users.
• The model file is downloaded once and cached. Switching models is just a path change — no re-download needed if already present.
• OpenAI API (set OPENAI_API_KEY) is even more accurate than large and handles technical vocabulary better.
• Free users get base model automatically. The cap (20/day) and model lock are independent — upgrading to small/medium/large requires a Pro license.

Troubleshooting

If the model file is not found, check the path. The default pattern is ~/.local/bin/ggml-{size}.bin. The WHISPER_MODEL env var overrides all other settings.

Ready to try it?

Get started free →