Diarization Index at a glance
Artificial Analysis launched the Diarization Index on November 13, 2026, a speech-to-text benchmark for speaker diarization. As of November 13, 2026, NVIDIA's Nemotron 3.5 Transcribe ranks first at 3.3% Word Diarization Error Rate across 97 conversations.
Ask your AI about this story
Use this story in your business: Artificial Analysis Diarization Index ranks speaker diarization: Nemotron 3.5 Transcribe leads at 3.3% WDER (Artificial Analysis). Ask your AI to explain what happened, how it could affect your business and how to put it to work. It also weighs price, speed, intelligence and the lab behind it. The link to the story is included.
What changed with Diarization Index
Artificial Analysis published the Diarization Index on November 13, 2026, a new speech-to-text benchmark that scores how well transcription APIs perform speaker diarization: assigning every word to the person who said it. The index measures Word Diarization Error Rate (WDER), the share of scored words given to the wrong speaker or missed entirely, across 97 held-out English conversations of 2 to 8 people (about 6.8 hours in total). Each API is called with diarization switched on and is never told how many people are talking; the index averages the 2, 3–4 and 5–8 speaker bands with equal weight.
As of November 13, 2026, NVIDIA's Nemotron 3.5 Transcribe leads the 23 ranked APIs at 3.3% WDER, less than half the 7.5% of second-placed Voxtral Small Transcribe 3 from Mistral. AssemblyAI's Universal (8.6%), Meta's Muse Voice Transcribe (8.6%) and Microsoft's MAI-Transcribe-2 (8.8%) follow, while OpenAI's GPT-4o Transcribe Diarize scores 10.4%, Google's Gemini 3.5 Transcribe 14.2% and Deepgram's Nova-3 16.7%. Nemotron 3.5 Transcribe is NVIDIA's open-weights transcription model, and Artificial Analysis lists it at $2 per 1,000 minutes of audio, against $4 for Voxtral Small Transcribe 3.
Diarization Index in numbers
Values are copied from the linked source on the date shown. They are not IA Compa measurements.
- Word Diarization Error Rate (WDER), Diarization Index · Nemotron 3.5 Transcribeas of · Source for Word Diarization Error Rate (WDER), Diarization Index: artificialanalysis.ai (opens in a new tab)
- 3.3%
- Word Diarization Error Rate (WDER), Diarization Index · Voxtral Small Transcribe 3as of · Source for Word Diarization Error Rate (WDER), Diarization Index: artificialanalysis.ai (opens in a new tab)
- 7.5%
- Transcription price per 1,000 minutes of audio · Nemotron 3.5 Transcribeas of · Source for Transcription price per 1,000 minutes of audio: artificialanalysis.ai (opens in a new tab)
- $2
- Transcription price per 1,000 minutes of audio · Voxtral Small Transcribe 3as of · Source for Transcription price per 1,000 minutes of audio: artificialanalysis.ai (opens in a new tab)
- $4
Why Diarization Index matters for businesses
Why it matters: diarization is what turns a transcript into a business record — who promised what on a sales call, which agent said what in a support recording, who owns the action item in a meeting. At 3.3% WDER, about 1 word in 30 is attached to the wrong person; at 7.5% it is about 1 in 13, and the weakest ranked API misattributes about 1 in 5. Businesses buying transcription for call-center quality checks, meeting notes or voice-agent workflows can now compare that accuracy directly, API by API — and the current leader is also among the cheapest per hour of audio.
Put it to work: AI services
Get AI News by email
Verified AI stories, each with its official source, on Monday and Thursday mornings. Add breaking news to hear about the biggest ones as soon as we publish them.
Every story links its official source
Free, in English or Spanish
Unsubscribe with one click, any time
Sources for this story
- Speech to Text (ASR) Providers Leaderboard & Comparison — Diarization Index (opens in a new tab)Official
Artificial Analysis · artificialanalysis.ai
- Speech to Text Benchmarking Methodology — Diarization Evaluation (opens in a new tab)Official
Artificial Analysis · artificialanalysis.ai
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #speaker-diarization
- #diarization-index
- #word-diarization-error-rate
- #speech-to-text
- #nemotron-3-5-transcribe
- #voxtral
- #artificial-analysis
- #benchmarks
