The short answer
Phonon-2.5 averages 11.71 % word error across the 25 FLEURS languages, against 12.18 % for Parakeet Redux, a model of about the same size, and 13.26 % for the 2.5 GB Parakeet TDT 0.6B v3. Phonon-2.5 is ahead of Redux in 14 of the 25 languages, and ahead of Parakeet TDT 0.6B v3 in 14.
The base model stays ahead on several large languages (French, German, Italian, Dutch, Polish, Russian), and Phonon-2.5 is ahead on most of the smaller ones, by wide margins in Greek, Latvian, Slovenian and Lithuanian. If your audio is in one language, the per-language row matters more than the average.
Per-language FLEURS word error (%, lower is better)
Full FLEURS test set of each language, the same code and normalisation for all three models. Sizes: Phonon-2.5 156 MB, Parakeet Redux 178 MB, Parakeet TDT 0.6B v3 2,508 MB.
| Language | Phonon-2.5 | Parakeet Redux | Parakeet TDT 0.6B v3 |
|---|---|---|---|
| English | 6.58 | 7.32 | 6.53 |
| Spanish | 4.80 | 4.99 | 4.69 |
| French | 8.12 | 9.03 | 6.61 |
| German | 6.67 | 7.57 | 5.96 |
| Italian | 5.21 | 5.12 | 4.75 |
| Portuguese | 6.53 | 6.44 | 6.27 |
| Dutch | 9.21 | 9.45 | 7.93 |
| Polish | 10.20 | 10.19 | 8.55 |
| Russian | 8.67 | 9.01 | 7.21 |
| Ukrainian | 7.89 | 8.05 | 7.27 |
| Romanian | 11.88 | 11.89 | 13.13 |
| Czech | 10.40 | 11.41 | 11.93 |
| Slovak | 8.89 | 8.17 | 9.72 |
| Bulgarian | 13.34 | 13.02 | 13.48 |
| Croatian | 12.24 | 11.43 | 13.21 |
| Swedish | 14.53 | 14.70 | 15.44 |
| Danish | 18.20 | 17.91 | 19.05 |
| Finnish | 12.02 | 11.84 | 13.08 |
| Estonian | 11.91 | 10.29 | 14.34 |
| Latvian | 14.92 | 14.76 | 23.66 |
| Lithuanian | 17.94 | 18.05 | 21.68 |
| Hungarian | 15.45 | 15.81 | 15.38 |
| Slovenian | 18.53 | 18.88 | 24.70 |
| Maltese | 17.91 | 16.02 | 20.86 |
| Greek | 20.83 | 33.25 | 36.11 |
| Average | 11.71 | 12.18 | 13.26 |
English
On the seven English test sets of the Open ASR Leaderboard, scored with the leaderboard's code on the full sets, Phonon-2.5 English (158 MB) averages 4.82 %, the 25-language Phonon-2.5 5.06 %, Parakeet TDT 0.6B v3 4.96 % (its published row) and Parakeet Redux 5.69 %.
| Model | Size | LS clean | LS other | AMI | Earnings-22 | GigaSpeech | SPGISpeech | VoxPopuli | Average |
|---|---|---|---|---|---|---|---|---|---|
| Phonon-2.5 English | 158 MB | 1.64 | 3.56 | 8.76 | 6.22 | 7.99 | 3.31 | 2.28 | 4.82 |
| Phonon-2.5 (25 languages) | 156 MB | 1.60 | 3.41 | 9.12 | 6.72 | 8.04 | 3.31 | 3.23 | 5.06 |
| Phonon-2 | 164 MB | 1.72 | 3.92 | 9.37 | 6.96 | 8.35 | 3.70 | 2.46 | 5.21 |
| Underdog Bark 3.0 | 163 MB | 1.89 | 4.08 | 7.82 | 7.05 | 7.95 | 3.72 | 2.05 | 4.94 |
| Parakeet TDT 0.6B v3 | 2,508 MB | 1.52 | 3.13 | 9.42 | 5.85 | 7.99 | 3.63 | 3.19 | 4.96 |
| Parakeet Redux | 178 MB | 1.94 | 4.35 | 9.16 | 7.90 | 8.62 | 4.01 | 3.87 | 5.69 |
What is the same, what is different
- The same 25 languages, so one table compares like with like.
- Phonon-2.5 is 16 times smaller than Parakeet TDT 0.6B v3 and loads in a fraction of the memory, which is what makes it fit on a phone or in a small container.
- Phonon-2.5 runs through the fermion runtime (pip, Docker, MLX); Parakeet TDT 0.6B v3 runs through NVIDIA NeMo or Transformers. The Parakeet Core ML and FluidAudio paths serve Parakeet on Apple devices; Phonon-2 has its own Core ML package for the Neural Engine.
Questions
- Is Phonon-2.5 more accurate than Parakeet TDT 0.6B v3?
- On average across the 25 FLEURS languages, yes: 11.71 % against 13.26 % word error, and ahead in 14 of 25 languages. The base is ahead on several large languages such as French, German and Russian.
- What is Parakeet Redux?
- A 178 MB speech recognition model for the same 25 languages. On FLEURS it averages 12.18 % against Phonon-2.5's 11.71 %, and on the seven English sets 5.69 % against 5.06 %.
- Which model should I use for English only?
- Phonon-2.5 English, 158 MB, averages 4.82 % on the seven Open ASR Leaderboard sets, better than Phonon-2.5, Parakeet TDT 0.6B v3 and Parakeet Redux on that average.
- Do the Parakeet Core ML runners work with Phonon-2.5?
- No. Phonon-2 ships its own Core ML package for the Apple Neural Engine (FermionResearch/Phonon-2-CoreML). Phonon-2.5 runs on Apple silicon through the MLX engine in the fermion runtime.
Related
Sources and method
- All FLEURS and English numbers: the Phonon-2.5 release post tables (components/site/phonon25-data.ts), 2026-10-22; Parakeet TDT 0.6B v3 English row is NVIDIA's published row.
- Model sizes: Hugging Face file sizes, read 2026-10-22.
Canonical: https://www.fermionresearch.com/models/phonon-2-5/vs-parakeet/