Transcription through OpenRouter and other OpenAI-compatible servers
Also released that day: v0.36.1
- #269 d664060 Transcription through OpenRouter and other OpenAI-compatible servers. The compatible provider (OPENAI_COMPATIBLE_*) now serves transcription models as well as language models, so TRANSCRIPTION_MODEL=openrouter:openai/whisper-large-v3-turbo works next to AI_MODEL=openrouter:…. Whether a transcription model returns the word timestamps that cuts and captions need is no longer decided from its name: the first time Reelsaw meets a model it transcribes a 5-second test clip and checks the answer, remembers the result for 30 days, and refuses a model that fails with that reason. Test under Settings → Models runs the clip again, as do the new test_model MCP tool and reelsaw test-model command, and the model picker marks which models passed. The picker also lists OpenRouter's speech models and the AI Gateway's transcription models from what those services report.