Cohere Transcribe
A Tier · 8.0/10
Cohere's first audio model -- launched 2026-03-26 under Apache 2.0, 2B parameters, #1 on Hugging Face Open ASR Leaderboard (5.42 avg WER), 14 enterprise-critical languages. Free API with rate limits; Model Vault for production
Score Breakdown
The Good and the Bad
What we like
- +#1 on Hugging Face Open ASR Leaderboard as of 2026-03-26 -- 5.42 average WER across 8 English benchmarks, beating IBM Granite 4.0 1B Speech (5.52), NVIDIA Canary Qwen 2.5B (5.63), ElevenLabs Scribe v2 (5.83), and OpenAI Whisper Large v3 (7.44)
- +Apache 2.0 open weights mean you can self-host without license fees -- rare for production-grade ASR at this accuracy tier. vLLM integration was contributed upstream so production serving works out of the box
- +Encoder-heavy architecture (90%+ of parameters in encoder, lightweight decoder) is explicitly optimized for serving efficiency -- offline throughput is 3x higher than comparable open-source models at similar WER
- +Trained on 14 enterprise-critical languages: English, German, French, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Arabic, Vietnamese, Mandarin, Japanese, Korean -- positioned squarely for European and APAC enterprise transcription
What could be better
- −No speaker diarization, word-level timestamps, or streaming WebSocket API -- this is a transcription-only model, not a full ASR platform like Grok Speech or Deepgram. Build those features yourself
- −Trained to expect a language tag and monolingual audio -- code-switched (En, X) transcription works sometimes but is not a first-class use case
- −Like most AED speech models, 'eager to transcribe' -- background noise and silence can turn into hallucinations without a VAD (voice activity detection) front-end
- −Text-to-speech is NOT in this release -- Cohere Transcribe is STT only. If you need both, pair with Grok TTS, ElevenLabs, or Microsoft MAI-Voice-1
Pricing
Open-weights (Apache 2.0)
- ✓Self-host the 2B-parameter model from Hugging Face
- ✓Commercial use allowed, no license fee
- ✓vLLM production-serving support (merged upstream)
- ✓3x higher offline throughput vs. similarly-sized models
Cohere API (free tier)
- ✓Hosted inference for low-setup experimentation
- ✓Rate-limited (see Cohere docs)
- ✓No commercial SLA
Cohere Model Vault (production)
- ✓Dedicated deployment, no rate limits
- ✓Per-hour instance pricing
- ✓Discounted plans for longer-term commitments
- ✓Cohere North enterprise platform integration
Known Issues
- OWNERSHIP-CHANGE WATCH (2026-04-24): Cohere announced it is 'joining forces' with Aleph Alpha. Vendor blog (cohere.com/blog/cohere-alephalpha-join-forces) frames this as a merger / strategic combination rather than a clean acquisition, backed by Schwarz Group as lead investor with EUR 500M (USD 600M) Series E structured financing. STACKIT (Schwarz Digits' sovereign cloud) will serve as the 'technical backbone of this transatlantic AI initiative', targeting regulated sectors (public sector, finance, defense, energy, manufacturing, telecom, healthcare). Effective date / closing date not specified; deal announced as planned/pending. No immediate product or pricing changes for Cohere Transcribe customers, but a future combined-entity roadmap may reshape model lineup, support guarantees, and pricing -- worth tracking before any new long-term Cohere Model Vault commitment.Source: Cohere blog: Cohere and Aleph Alpha join forces (cohere.com/blog/cohere-alephalpha-join-forces), TechCrunch, CNBC · 2026-04-24
- No speaker diarization or word-level timestamps at launch. If your product needs multi-speaker separation (call centers, meetings) you'll need to add a diarization model on top or choose Grok STT / Deepgram / AssemblyAISource: Cohere Hugging Face blog · 2026-03
- Model 'eagerly' transcribes non-speech audio -- prepend a VAD or noise gate for production use to avoid hallucinations on floor noiseSource: Cohere Labs limitations note · 2026-03
- Only 14 languages supported -- narrower than Whisper Large v3 (99 languages) or Grok STT (25+). For rare-language workloads, open-source Whisper or proprietary APIs still winSource: Cohere Labs model card · 2026-03
Best for
Enterprise teams transcribing English, European, and major APAC languages at scale who want open weights they can self-host, fine-tune, or deploy on-prem. The Apache 2.0 license removes a major procurement blocker compared to proprietary ASR, and the accuracy tier is now best-in-class for open models.
Not for
Call-center and meeting-transcription products that need speaker diarization and streaming out of the box -- Grok STT, Deepgram, or AssemblyAI ship those features. Also not ideal for rare-language workloads (sub-Saharan African languages, indigenous languages) where Whisper Large's 99-language coverage is still hard to beat.
Our Verdict
Cohere Transcribe is the first time an open-weights ASR model has genuinely topped the Open ASR Leaderboard in head-to-head benchmark terms, and the Apache 2.0 license makes it a real procurement unlock for enterprises that couldn't self-host Whisper at the accuracy tier they needed. The limitations (no diarization, no streaming, no TTS counterpart) mean it won't replace full ASR platforms for call-center or meeting products -- but for batch enterprise transcription (document pipelines, media indexing, compliance recording) it's now the default open-source pick. Pair with Grok TTS or ElevenLabs on the output side and you have a fully-open voice stack for the first time.
Sources
- Cohere Labs: Introducing Cohere-transcribe (accessed 2026-04-18)
- TechCrunch: Cohere open-source voice transcription (accessed 2026-04-18)
- MarkTechPost coverage (accessed 2026-04-18)
- Hugging Face Open ASR Leaderboard (accessed 2026-04-18)
Explore more Cohere Transcribe rankings
Deeper leaderboards, benchmarks, task-specific tier lists, and status/pricing pages for Cohere Transcribe.
The Tier List Tuesday
Weekly newsletter: tier movers, new entrants, and the VS of the week. Built from our daily AI-tool sweeps. No spam, unsubscribe anytime.
Alternatives to Cohere Transcribe
ElevenLabs
**SFX v3 (2026-10-29), the newest ElevenLabs sound-effects model, follows prompts that describe several sounds more closely and handles music and vocals better than SFX v2; it is in ElevenCreative (Sound Effects tab, Flows, Studio, Music Editor, mobile app) and in the API as eleven_text_to_sound_v3 on the same endpoint, though looping still needs the v2 model.** **Eleven Realtime (2026-10-20) adds two speech-to-speech agent models, v1 Mini and v1 Max, in ElevenAgents and an OpenAI Realtime-compatible API at $0.08 per minute of connected audio plus LLM tokens; Realtime Translation followed on 10/21 in alpha for API enterprise customers.** **Instant Soundtracks (2026-10-15) scores any uploaded video with original music and timed sound effects in ElevenCreative, ElevenMusic and the API, with effects on separate layers and the option to use the clip's existing audio as a style reference.** **ElevenAgents gained two new tools in the week of 2026-10-06: Architect (alpha, 10/06), a built-in expert that diagnoses agent failures from transcripts and drafts tested fixes, and synthetic voice detection (10/09), which flags AI callers in the first seconds of a call -- enterprise-only for now. ElevenCreative added Brand Kit (10/09).** **Two days after the v4 launch, ElevenLabs closed a $300M employee tender at a $22B valuation (2026-09-30), double its February Series D, with enterprise now 55% of revenue -- and the pricing page is running v4 launch promos (extra v4 credits on Creator plans and above, see the Oct 2 note).** **Eleven v4 and Eleven v4 Turbo launched 2026-09-28 -- a fourth-generation voice model on an entirely new architecture**, ranked #1 on Artificial Analysis' Provider Voice Arena and preferred by ~75% of listeners in blind head-to-head tests against Cartesia Sonic 3.6, Inworld TTS-2 and Google's Gemini 3.8 Flash and Flash-Lite TTS; v4 Turbo answers with a ~100ms median inference latency (~150ms to first speech over WebSocket) and is co-optimized with ElevenAgents. Both cover 90+ languages, follow inline audio tags like [laughs] or [phone buzzing] more accurately, clone voices with better speaker similarity, and are live now in ElevenAgents, ElevenCreative and the API (Text to Dialogue). Eleven v3 becomes the previous generation. The 2026-09-10 Universal Music Group licensing deal and CLI v1 (8/24) still stand.
Murf AI
Text-to-speech that actually sounds like a real person read your script -- not a robot trying its best
Descript
Edit audio and video by editing text -- the 'Google Docs of media editing' actually lives up to the hype
Speechify
Text-to-speech reader that turns articles, docs, and PDFs into natural-sounding audio
Microsoft MAI-Voice-2
Microsoft's in-house expressive TTS line. **MAI-Voice-2.1 and MAI-Voice-2.1-Flash launched 2026-10-01: 23 languages and 26 locales with one voice that keeps the same speaker and a native accent across languages, voice cloning from a few seconds of reference audio with consent guardrails, priced at $22 per 1M characters for 2.1 and $15 per 1M characters for Flash, which generates 45 seconds of audio at 150 ms end-to-end latency and is pitched at voice agents alongside MAI-Transcribe-2-Streaming.** These supersede **MAI-Voice-2 (Build, 2026-06-02)**: 15 languages, emotion tags, zero-shot cloning from a 5-60 s clip, preferred over MAI-Voice-1 72% of the time. Microsoft also shipped a MAI-Voice-2-Flash and MAI-Image-2.5-Pro in July that this page had not recorded -- see the Oct 2 note.
Grok Speech (STT + TTS APIs)
**Grok Voice 3.0, xAI's new speech-to-speech model, shipped 2026-10-14 at the same $0.08 per minute as 2.0, which xAI's rate card now flags as on its way out; grok-voice-latest switched over automatically, and xAI cites a #1 score of 86.9 on Artificial Analysis' Speech-to-Speech Index.** xAI's standalone voice APIs. **Grok Voice Transcribe 2.0 shipped 2026-09-18** -- xAI calls it twice as accurate as 1.0 at the same $0.10/hr batch and $0.20/hr streaming, ranks it first for accuracy among 32 streaming models on Artificial Analysis, and has already made it the default STT model; **1.0 is being retired in the coming weeks, so pin grok-voice-transcribe-1.0 if you need it**. **PRICE CORRECTION: text-to-speech is $15.00 per 1M characters on xAI's live rate card (verified 2026-09-21), not the $4.20 this page carried since April.** Speech-to-speech agent $0.08/min, 26 flagship voices, ~1-minute voice cloning, no-code Voice Agent Builder
GPT-Live (ChatGPT Voice)
**GPT-Live-1.1 and a smaller GPT-Live-1.1 Quip began rolling out in ChatGPT Voice on 2026-10-15, with more natural accents, shorter spoken cues while it searches, and new voices for French, German, Hindi, Japanese, Korean and Mexican Spanish plus an English voice called Bubbie.** OpenAI's full-duplex voice models (launched in ChatGPT 2026-07-08) -- listens and speaks at the same time, backchannels naturally, and delegates hard questions to a backend model while keeping the conversation going. **GPT-Live-1 reached the API on 2026-09-10 at $0.05 per minute** for the voice layer, with telephony support, keyword biasing, delegation to any backend model (GPT-6 Astra, Luna, or a third-party model) and a wider voice set. In ChatGPT: GPT-Live-1 for paid tiers, GPT-Live-1 mini for Free. **Five days after the API launch, Google priced Gemini 3.8 Live at $0.005/min audio in and $0.018/min audio out (2026-09-15)** -- a fraction of GPT-Live-1's $0.05/min voice layer, though each stack bills backend or text tokens on top