Developers who want an accurate open speech-to-text model they can host themselves get it free under Apache 2.0, though it covers only 14 languages.
Kenning's take
What it does
Cohere Transcribe is a 2-billion-parameter speech-to-text model released as open source under Apache 2.0. It covers 14 languages including English, French, German, Arabic, Chinese and Japanese, and at launch ranked first for accuracy on the Hugging Face Open ASR Leaderboard. It can be downloaded from Hugging Face, tried free through the Cohere API with rate limits, or run on Cohere Model Vault.
- Open weights under Apache 2.0
- 14 supported languages
- Up to 25MB audio files
- Conformer encoder with a small Transformer decoder
- No speaker labels, timestamps or automatic language detection
Who it suits. Developers and companies building transcription, meeting notes or voice support systems who want a model they can run themselves.
Key facts
- Category
- AI models
- Made by
- Cohere
- Based in
- Canada
- Released
- 26 March 2026
- How to get it
- Free download from Hugging Face; free rate-limited use through the Cohere API; paid dedicated Model Vault for production
- Works on
- API, Self-hosted (open weights)
- Free plan
- Yes: open weights on Hugging Face, and free rate-limited API access
- API price
- Free for rate-limited experimentation (5 requests per minute on trial keys); production through Model Vault
Pricing
| Plan | Price | Billed |
|---|---|---|
| Model Vault | $contact sales | hour per instance |
FAQ
When was Cohere Transcribe released?
Cohere announced Transcribe on 26 March 2026.
Is Cohere Transcribe free?
Yes. The weights are open source under Apache 2.0, and the Cohere API offers free rate-limited use. Production use without limits needs a paid Model Vault.
Which languages does Cohere Transcribe support?
English, German, French, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Vietnamese, Chinese, Arabic, Japanese and Korean.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.