All tasks / Comparisons
MAI-Transcribe-2 (preview) vs StepAudio 3 ASR
Compared only on public benchmark results, across the 1 task where both are listed: they share a tier in 1.
MAI-Transcribe-2 (preview)
Microsoft AI's in-house speech recognition models for 60 languages, offered through Microsoft Foundry and Azure Speech.
- Vendor
- Microsoft AI
- Weights
- Closed: API and apps only
- Released
- 3 Sep 2026
- API price
- $0.0017 per minute of audio (Launch rate $0.10 per hour of audio (introductory through 2026-12-31 per secondary reports) divided by 60.)
- License
- proprietary
- Last verified
- 18 Oct 2026
StepAudio 3 ASR
StepFun's closed speech recognition model served through the StepFun API platform.
- Vendor
- StepFun
- Weights
- Closed: API and apps only
- Released
- 14 Sep 2026
- API price
- $0.0067 per minute of audio (Artificial Analysis lists $6.67 per 1,000 minutes (StepAudio 3 ASR); divided by 1,000.)
- License
- proprietary
- Last verified
- 19 Oct 2026
Task by task
| Task | MAI-Transcribe-2 (preview) | StepAudio 3 ASR | Higher tier |
|---|---|---|---|
| Transcription Models | S2% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming) · Cloud | S2.5% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming) · Cloud | Same tier |
A tier spans a range of scores, and small gaps on a leaderboard are often within its margin of error, so only a different tier counts as ahead. How we rank.
Same benchmark, both results
| Benchmark | MAI-Transcribe-2 (preview) | StepAudio 3 ASR | Snapshot |
|---|---|---|---|
| Artificial Analysis Speech to Text: AA-WER v2 (non-streaming) AA-WER v2: word error rate across AA's own test set plus cleaned VoxPopuli and Earnings-22 subsets (non-streaming), lower is better | 2% | 2.5% | 19 Oct 2026 |