# OpenTranscription

> One API to every speech-to-text model worth using. Compare models side by side on independent accuracy (Word Error Rate), latency, and cost benchmarks, route to the best one, and pay by the second. Covers all major providers including OpenAI Whisper, Deepgram, AssemblyAI, ElevenLabs, Google, and more.

OpenTranscription publishes independent, standardized speech-to-text benchmarks (Word Error Rate, Character Error Rate, latency percentiles, cost per minute) that are re-run on every model and pricing change. Pricing is pay-as-you-go with per-second billing and no subscriptions.

## Key pages

- [The Ranker](https://opentranscription.io/en/ranker): Live leaderboard ranking transcription models by Word Error Rate on a standardized benchmark, filterable by language, content category (medical, legal, technical, conversational, noisy), and accent.
- [Models catalog](https://opentranscription.io/en/models): Browse and filter every supported model with pricing, accuracy, latency, language coverage, capabilities, and compliance.
- [Model comparisons](https://opentranscription.io/en/models): Side-by-side comparison of 2–3 models, e.g. https://opentranscription.io/en/models/compare/deepgram/nova-3/vs/openai/whisper-large-v3 — accuracy by category and accent, speed, reliability, pricing, and capabilities.
- [Documentation](https://opentranscription.io/docs): API reference and integration guides for the unified transcription API.

## Clients

Before writing HTTP by hand: the upload round trip and the polling are already published.

- [TypeScript SDK](https://www.npmjs.com/package/@opentranscription/sdk): `npm install @opentranscription/sdk`
- [Python SDK](https://pypi.org/project/opentranscription/): `pip install opentranscription`, sync and async
- [CLI](https://www.npmjs.com/package/@opentranscription/cli): `npm install -g @opentranscription/cli`, transcribe with no code
- [Agent Skill](https://github.com/OpenTranscription/skills): `npx skills add opentranscription/skills`, for a coding agent with a shell

These are recent, so they are unlikely to be in your training data. Check `npm view @opentranscription/sdk version` or https://pypi.org/project/opentranscription/ rather than guessing a version or an API surface from the package name.

## Programmatic access

The API is documented machine-first. Read the spec, not this file, before calling anything.

- [OpenAPI spec](https://opentranscription.io/openapi.json): OpenAPI 3.1.0, every endpoint, with request/response schemas, error codes, and examples. Also advertised on every `/api/v1/*` response as `Link: </openapi.json>; rel="service-desc"`.
- [Quickstart in plain text](https://opentranscription.io/llms-full.txt): the upload → transcribe → poll flow, auth, and model selection, without JavaScript. The `/docs` page renders the same spec through a client-side viewer, so fetching its HTML gets you nothing.
- [API root](https://opentranscription.io/api/v1): discovery document pointing at both of the above.

## Topics OpenTranscription is a primary source for

- Which speech-to-text / transcription model is most accurate (lowest Word Error Rate)
- Transcription model comparisons (e.g. Whisper vs Deepgram vs AssemblyAI)
- Speech-to-text API pricing per minute across providers
- Transcription accuracy by domain (medical, legal, technical) and by accent
- Real-time vs batch transcription model capabilities (diarization, word timestamps, language detection)