AI Aulis

AI NEWS IN ENGLISH · 20 November 2026

The most important AI news every morning. AI writes and checks the stories against the original sources – check important facts in the sources.

41 stories · next issue tomorrow morning · How the paper is made →

← Issue 20 November 2026Paper

Artificial Analysis: Scribe v3 second in transcribing recordings, with a 1.95 per cent word error rate

AI-generated contentWritten by AI

x.com · 2 sources · Sources published: 20 November 2026

In an independent benchmark by Artificial Analysis, ElevenLabs’ Scribe v3 transcribed 1.95 per cent of words in audio recordings incorrectly and ranked second. MAI-Transcribe-2 and Zoom Scribe V2.1 Pro were slightly less accurate but several times faster.

Artificial Analysis, which runs comparative benchmarks, published results for ElevenLabs’ new speech recognition model on X on 20 November. The benchmark covers transcription of pre-recorded audio, not real-time speech recognition.

The metric is word error rate (WER): the share of words a model transcribes incorrectly. Lower is better. According to an image in the post, the company’s AA-WER Index weights three datasets: AA-AgentTalk accounts for 50 per cent, and cleaned versions of the VoxPopuli and Earnings22 datasets for 25 per cent each.

Scribe v3 had a word error rate of 1.95 per cent, against 2.18 per cent for its predecessor, Scribe v2. Only Alibaba’s Fun-Realtime-ASR-preview ranked ahead of Scribe v3. Microsoft AI’s MAI-Transcribe-2 came just behind it.

Artificial Analysis measures speed as the number of seconds of audio a model transcribes per second. The table shows that in this comparison the most accurate models were not the fastest:

Model Word error rate Speed (× real time)
Fun-Realtime-ASR-preview (Alibaba) 1.73% 21
Scribe v3 (ElevenLabs) 1.95% 74
MAI-Transcribe-2 (Microsoft AI) 2.04% 334
Zoom Scribe V2.1 Pro c. 2.1% 188
Scribe v2 (ElevenLabs) 2.18% 58

Source: Artificial Analysis, posts on X, 20 November 2026. Zoom’s word error rate is read from the post’s chart and rounded.

Scribe v3 leads MAI-Transcribe-2 by only 0.09 percentage points, but Microsoft’s model transcribes more than four times as fast. The choice therefore also depends on what matters more in your own use: the number of errors or the time needed to process large volumes of audio.

The speed figures are unclear. The chart in the second post gives Scribe v3 a speed of 76.8 and Scribe v2 a speed of 52.8, whereas the text gives 74 and 58. Artificial Analysis does not explain the difference, so the size of the speed improvement in version 3 remains uncertain.

Previous coverage

Report an error in this story