Side by side
See the difference.
Scores, capabilities and costs for the models you’re considering.
Scroll to compare models →
| 3 models | fal-aiMiniMax Speech 2.8 [HD]Remove × | ideogramIdeogram V4.5 EditRemove × | bosonaibosonai/higgs-audio-v3-sttRemove × |
|---|---|---|---|
| What’s different | Use timed pause tags, a custom pronunciation dictionary, and separate pitch, energy and timbre controls for narration. | Mask-guided edits with a high-precision mode that restores unchanged pixels; combines a source image with extra references. | 2.68B parameters · 1 language · Encoder: Whisper · Decoder: Qwen3 |
| Date | Feb 14, 2026Released | Sep 30, 2026Listed | Oct 17, 2026Discovered |
API prices USD · provider configuration shown below | |||
| Listed rate | $100.00 in / — out/ 1M tokensReplicate · all routes ↗input tokens; not characters | See provider quotes | See provider quotes |
| Provider endpoints | 2 endpoints → | 0 endpoints → | 0 endpoints → |
Benchmarks Each result retains its evaluation configuration | |||
| AA Native Voice ↗Unversioned · provider native voices | 1,174.07 ↗Provider native voices · all categories · US + UK accents · 8 listed voicesIndependent result95% CI 1,164.07–1,184.07 · n=5,194 | Not reported | Not reported |
| AMI-Cleaned WER ↗English short-form; normalization unversioned | Not reported | Not reported | 7.19% WER ↗bosonai/higgs-audio-v3-sttLeaderboard entry |
| Earnings22-Cleaned-AA-chunked WER ↗English short-form; normalization unversioned | Not reported | Not reported | 7.04% WER ↗bosonai/higgs-audio-v3-sttLeaderboard entry |
| Gigaspeech-Cleaned WER ↗English short-form; normalization unversioned | Not reported | Not reported | 7.45% WER ↗bosonai/higgs-audio-v3-sttLeaderboard entry |
| LS Clean WER ↗English short-form; normalization unversioned | Not reported | Not reported | 1.08% WER ↗bosonai/higgs-audio-v3-sttLeaderboard entry |
| LS Other WER ↗English short-form; normalization unversioned | Not reported | Not reported | 2.61% WER ↗bosonai/higgs-audio-v3-sttLeaderboard entry |
| SPGISpeech WER ↗English short-form; normalization unversioned | Not reported | Not reported | 2.07% WER ↗bosonai/higgs-audio-v3-sttLeaderboard entry |
| URGENT2024 WER ↗English short-form; normalization unversioned | Not reported | Not reported | 11.88% WER ↗bosonai/higgs-audio-v3-sttLeaderboard entry |
| Voice Arena Monsoon WER ↗English short-form; normalization unversioned | Not reported | Not reported | 3.97% WER ↗bosonai/higgs-audio-v3-sttLeaderboard entry |
| Voxpopuli-AA-Cleaned WER ↗English short-form; normalization unversioned | Not reported | Not reported | 3.73% WER ↗bosonai/higgs-audio-v3-sttLeaderboard entry |
Specifications | |||
| Context window | — | — | — |
| Max output | — | — | — |
| Parameters | — | — | 2.68B |
| Audio input | Not reported | Not reported | Supported |
| Transcription | Not reported | Not reported | Supported |