IBMFermion ResearchAudio matchupSpeech to text
Granite Speech 5.0 470M TurboCTCvs.Phonon-2
Which model is better?
Granite Speech 5.0 470M TurboCTC is slightly ahead in independent tests, but the lead is not certain.
Pick Granite Speech 5.0 470M TurboCTC for
Pick Phonon-2 for
Shared tests
Independent evaluators and leaderboards only, the newest shared snapshot per evaluator. With 95 % ranges a lead counts when they do not overlap; for hit rates, when it holds in a 95 % test. Without any uncertainty figure at most “slightly ahead”.
Hugging Face Open ASR
Granite Speech 5.0 470M TurboCTC slightly aheadGranite Speech 5.0 470M TurboCTC is clearly ahead in 0 of 9 subtests and slightly in 6, Phonon-2 in 0 clearly and slightly in 2. 1 is level.
Open ASR English short-form WER: Voxpopuli-AA-Cleaned
lower is better, as of October 28, 2026
Open ASR English short-form WER: Private (scripted)
lower is better, as of October 28, 2026
Open ASR English short-form WER: URGENT24
lower is better, as of October 28, 2026
Show all 9 subtests
| Subtest | Granite Speech 5.0 470M TurboCTC | Phonon-2 | Verdict |
|---|---|---|---|
| Open ASR English short-form WER: Voxpopuli-AA-Cleaned (lower is better) | 4.9 % | 2.5 % | Phonon-2 slightly ahead |
| Open ASR English short-form WER: Private (scripted) (lower is better) | 4.1 % | 5.2 % | Granite Speech 5.0 470M TurboCTC slightly ahead |
| Open ASR English short-form WER: URGENT24 (lower is better) | 11.8 % | 14.9 % | Granite Speech 5.0 470M TurboCTC slightly ahead |
| Open ASR English short-form WER: AMI-Cleaned (lower is better) | 7.7 % | 9.3 % | Granite Speech 5.0 470M TurboCTC slightly ahead |
| Open ASR English short-form WER: Average WER (11-set macro-average) (lower is better) | 6.4 % | 6.9 % | Granite Speech 5.0 470M TurboCTC slightly ahead |
| Open ASR English short-form WER: Gigaspeech-Cleaned (lower is better) | 8.7 % | 8.4 % | Phonon-2 slightly ahead |
| Open ASR English short-form WER: Voice Arena Monsoon (lower is better) | 4.8 % | 4.9 % | Granite Speech 5.0 470M TurboCTC slightly ahead |
| Open ASR English short-form WER: Earnings22-Cleaned-AA-chunked (lower is better) | 6.8 % | 7.0 % | Granite Speech 5.0 470M TurboCTC slightly ahead |
| Open ASR English short-form WER: Private (conversational) (lower is better) | 14.0 % | 14.1 % | level |
Source · latest October 28, 2026
All data side by side
Prices are verbatim from the provider, because audio models bill per minute, character or token.
| Granite Speech 5.0 470M TurboCTC | Phonon-2 | |
|---|---|---|
| Task | Speech to text | Speech to text |
| Developer | IBM | Fermion Research |
| Status | Open weights | Open weights |
| Released | August 25, 2026 | n/a |
| Weights | open | open |
| Runs locally | yes | yes |
| Parameters | 0.47 B | n/a |
| License | Apache-2.0 | CC-BY-4.0 |
| Popularity (out of 100) | 52 | 44 |