whatmodel.sh / stt / Universal-3 Pro
Universal-3 Pro
↗ share
+ contribute
Universal-3 Pro is a closed-weights speech-to-text model from AssemblyAI, released February 2026. Licence: proprietary.
Catalogue identity notes
AssemblyAI speech_models=['universal-3-pro'] (leaderboard assemblyai_provider.py; short-form entry 'assemblyai/universal-3-pro', long-form/multilingual entry 'assembly/universal-3-pro'). Announced 2026-02-03. Distinct from catalogue assemblyai-universal-3-5-pro. AssemblyAI retired Universal-3 Pro (routing switch 2026-09-02; models page observed 2026-09-28 lists only universal-3-5-pro and universal-2), so api is null.
Score within the Open ASR English short-form WER (March 2026) comparison set only.
$ how it compares
The speech-to-text models ranked either side of Universal-3 Pro, out of 176 with a score, then more from AssemblyAI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 15 |  Granite Speech 5.0 470M TurboCTC NC | 64±4 | local | ||
| 16 |  Cohere Transcribe Arabic (07-2026) | 63±3 | local | ||
| 17 | Universal-3 Pro | 63±2 | comparison only | ||
| 18 | Granite Speech 4.1 2B | 62±3 | local | ||
| 19 | Granite 4.0 1B Speech | 62±3 | local | ||
| more from AssemblyAI | |||||
| 4 | Universal-3.6 Pro Realtime | 70±3 | api $0.0075/audio min | ||
| 8 | Universal-3.5 Pro | 67±3 | api $0.0035/audio min | ||
| 53 | Universal-Streaming (English) | 58±3 | api $0.0025/audio min | ||
see all speech-to-text models · find the ones that fit your machine
$ what it's good at
long-form transcription▁▁▅▁·▃▅█▆▁▃▂ #3 of 31 * · 13% of its score · 1 test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Open ASR English long-form WER | 6.8% ✎ | ▁▁▅▁·▃▅█▆▁▃▂ beats 28/30 | 13% |
multilingual transcription▂··▃▄█▆▅▄▃·▂ #3 of 37 * · 45% of its score · 1 test
Beats Universal-3.5 Pro, Cohere Transcribe Arabic (07-2026) here, though they rank higher overall. Trails SeamlessM4T v2 Large here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Open ASR multilingual WER | 3.2% ✎ | ▂··▂▃▂█▃█▂▃▁ beats 28/29 | 45% |
other tests4 not in a task · 41% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Open ASR English short-form WER (March 2026) | 6.2% ✎ | ▁▁▁▃··▁▁▂██▆ beats 76/86 | 13% |
| RW-Voice-EQ ASR · 4 real-world English sets, mean WER | 9.4% ✎ | █▃▃▅▃···▃··▃ beats 2/9 | 11% |
| VoiceCodeBench · batch canonical token/entity match (CTEM) | 84.5% ✎ | ▄···▄··▄▄▄·█ beats 1/6 | 10% |
| VoiceCodeBench · batch WER (300 English workplace segments) | 4.3% ✎ | ▄·▄·▄·▄█···▄ beats 6/6 | 7.3% |
$ where to run it
No verified API or download recorded yet · add one
$ how the score was worked out · Open ASR Leaderboard 71%, VoiceCodeBench (arXiv:2608.28… 17%, RW-Voice-EQ Bench (Hume AI) 11%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Open ASR Leaderboard | 3 | 71% |
 VoiceCodeBench (arXiv:2608.28916v2) · structured-token recovery ASR benchmark | 2 | 17% |
| RW-Voice-EQ Bench (Hume AI) | 1 | 11% |
| prior · pull toward the middle | 1% | |
| weighted average | = 62.6 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json