Side by side

See the difference.

Scores, capabilities and costs for the models you’re considering.

Comparison of FermionResearch/Phonon-2
1 modelFermionResearchFermionResearch/Phonon-2Remove ×
What’s different0.6B parameters · 1 language · Encoder: FastConformer (low-bit) · Decoder: TDT
DateOct 17, 2026Discovered
Listed rateSee provider quotes
Provider endpoints0 endpoints →
AMI-Cleaned WER ↗English short-form; normalization unversioned9.32% WER ↗FermionResearch/Phonon-2Leaderboard entry
Earnings22-Cleaned-AA-chunked WER ↗English short-form; normalization unversioned6.96% WER ↗FermionResearch/Phonon-2Leaderboard entry
Gigaspeech-Cleaned WER ↗English short-form; normalization unversioned8.35% WER ↗FermionResearch/Phonon-2Leaderboard entry
LS Clean WER ↗English short-form; normalization unversioned1.73% WER ↗FermionResearch/Phonon-2Leaderboard entry
LS Other WER ↗English short-form; normalization unversioned3.92% WER ↗FermionResearch/Phonon-2Leaderboard entry
SPGISpeech WER ↗English short-form; normalization unversioned3.7% WER ↗FermionResearch/Phonon-2Leaderboard entry
URGENT2024 WER ↗English short-form; normalization unversioned14.9% WER ↗FermionResearch/Phonon-2Leaderboard entry
Voice Arena Monsoon WER ↗English short-form; normalization unversioned4.91% WER ↗FermionResearch/Phonon-2Leaderboard entry
Voxpopuli-AA-Cleaned WER ↗English short-form; normalization unversioned2.46% WER ↗FermionResearch/Phonon-2Leaderboard entry
Context window—
Max output—
Parameters0.6B
Audio inputSupported
TranscriptionSupported

Add another model

Search all models →