Replicate (Whisper hosts)
Replicate's catalog of community-hosted Whisper variants behind one API.
Replicate's catalog of community-hosted Whisper variants behind one API.
Best for prototyping with specific Whisper or WhisperX builds without operating your own GPU. Pricing: per-second GPU compute.
What it is
Replicate hosts hundreds of community Whisper and WhisperX variants behind a single API. The official openai/whisper model and popular forks (incl. WhisperX with diarization) are runnable via a single REST call. Pricing is per-second GPU compute with model-specific hardware classes. Good for prototyping; for steady production volume an integrated vendor (Deepgram/AssemblyAI/Gladia) usually wins on cost and latency. Best fit: prototyping with specific whisper or whisperx builds without operating your own gpu. Caveats: cold-start latency; community models change ownership; slas are per-replicate, not per-model author. Feature flags from vendor docs: word-level timestamps. Directory tags: commercial-api, model-host. Last vendor-page check: 2026-05-12.
Watch out for: Cold-start latency; community models change ownership; SLAs are per-Replicate, not per-model author.
Install / use
replicate.run('openai/whisper:...', input={...})Features
| Speaker diarization | No |
| Word-level timestamps | Yes |
| Streaming / real-time | No |
| Languages supported | 99 |
| HIPAA eligible | No |
Replicate (Whisper hosts) vs Whipscribe
| Feature | Replicate (Whisper hosts) | Whipscribe |
|---|---|---|
| Category | Transcription APIs | Transcription APIs |
| Pricing | per-second GPU compute | free beta |
| Speaker diarization | — | Yes |
| Word timestamps | Yes | Yes |
| Streaming | No | No |
| Languages | 99 | 99 |
| Platforms | API | Web, API, MCP |
Alternatives to Replicate (Whisper hosts)
Whipscribe is a managed faster-whisper + whisperX service. If you want transcripts without running infrastructure, paste a URL or drop a file in the form below — you'll have a transcript in seconds.