Benchmarks › AA Speech to Text
AA Speech to Text
Transcribe audio and count word insertions, deletions and substitutions
1.7top score · Fun Realtime ASR Preview
51models ranked
9open-weights models
WERlower is better
Oct 20, 2026last updated
How the test works
1
What it is
Transcribe audio and count word insertions, deletions and substitutions
- Kinds of tasksSpeech recognition
- Number of items1,103 parent audio files (469 AA-AgentTalk + 628 VoxPopuli + 6 Earnings22)
2
Components
AA-AgentTalk 50%, VoxPopuli-Cleaned-AA 25%, Earnings22-Cleaned-AA 25%
- How runs combineAudio-duration-weighted WER within each dataset; 50% AA-AgentTalk + 25% VoxPopuli-Cleaned-AA + 25% Earnings22-Cleaned-AA.
- Agent / harnessDirect provider API; where supported: Transcribe the audio verbatim, outputting only spoken words in sequence
3
Score
WER
- Graded by(Substitutions + Insertions + Deletions) / reference words
- Better isLower
Sources: artificialanalysis.ai
AA Speech to Text leaderboardtop 10 of 51
Full leaderboard with filters →| Model | AA Speech to Text | Languages | Speaker diarization | Per audio minute | Price covers | Price | Released |
|---|---|---|---|---|---|---|---|
1.7 | — | — | — | — | |||
2 | en, es, fr +57 | Yes | $0.002 / min | MAI-Transcribe-2 | $0.002 / minute | Sep 2026 | |
2.1 | — | Yes | $0.004 / min | — | $0.004 / minute | Oct 2026 | |
2.2 | en, es, fr +97 | Yes | $0.004 / min | Exact native tariff; see source for billing scope. | $0.004 / minute | Jan 2026 | |
2.2 | — | Yes | $0.004 / min | — | $0.004 / minute | Oct 2026 | |
2.3 | en, es, fr +35 | Yes | $0.002 / min | Exact native tariff; see source for billing scope. | $0.002 / minute | Sep 2026 | |
2.4 | en, es, fr +40 | — | $0.006 / min | MAI-Transcribe-1.5 | $0.006 / minute | Jun 2026 | |
2.5 | — | — | — | — | |||
2.6 | en-GB, en-IN, en-US +80 | Yes | $0.005 / min | Exact native tariff; see source for billing scope. | $0.005 / minute | Aug 2026 | |
2.8 | — | — | $0.004 / min | Exact native tariff; see source for billing scope. | $0.004 / minute | Jul 2025 |
AA Speech to Text questions
What is AA Speech to Text?›
Transcribe audio and count word insertions, deletions and substitutions. It has 1,103 parent audio files (469 AA-AgentTalk + 628 VoxPopuli + 6 Earnings22). Created by Artificial Analysis.
Which model is best on AA Speech to Text?›
Fun Realtime ASR Preview leads AA Speech to Text with 2, ahead of MAI Transcribe 2 (2) and Voxtral Small Transcribe 3 (2).
What is the best open-weights model on AA Speech to Text?›
Whisper Large v2 with 4, then Canary Qwen 2.5B, NVIDIA (4).
How is AA Speech to Text scored?›
WER; (substitutions + insertions + deletions) / reference words.