---
title: "FLEURS results in 25 languages"
description: "Phonon-2.5 averages 11.71 % word error on FLEURS across 25 languages, against 12.18 % for Parakeet Redux and 13.26 % for Parakeet TDT 0.6B v3. Every language."
canonical: "https://www.fermionresearch.com/speech/fleurs-results-25-languages/"
source: "Fermion Research"
---

Speech recognition/facts checked 13 November 2026

# FLEURS results in 25 languages

The full per-language table, measured on the complete FLEURS test set of each language with one scoring code for all three models.

[Phonon-2.5 downloads and model card](https://www.fermionresearch.com/models/phonon-2-5/)Open weights on Hugging Face.

## The short answer

On the full FLEURS test sets of its 25 languages, Phonon-2.5 (156 MB) averages 11.71 % word error, against 12.18 % for Parakeet Redux (178 MB) and 13.26 % for NVIDIA's Parakeet TDT 0.6B v3 (2,508 MB). All three were scored with the same code on 2026-10-22.

The average hides a split. Parakeet TDT 0.6B v3 is lower than Phonon-2.5 in 11 languages, mostly the most widely spoken ones; Phonon-2.5 is lower in the other 14. Phonon-2.5 is lower than Parakeet Redux in 14 of the 25 and the lowest of the three in 6.

## Per-language word error rate

Word error rate, %, lower is better, on the full FLEURS test set of each language. Same scoring code and text normalisation for all three models; diacritics kept.

| Language | Phonon-2.5 (156 MB) | Parakeet Redux (178 MB) | Parakeet TDT 0.6B v3 (2,508 MB) | Lowest |
| --- | --- | --- | --- | --- |
| English | 6.58 | 7.32 | 6.53 | Parakeet TDT 0.6B v3 |
| Spanish | 4.80 | 4.99 | 4.69 | Parakeet TDT 0.6B v3 |
| French | 8.12 | 9.03 | 6.61 | Parakeet TDT 0.6B v3 |
| German | 6.67 | 7.57 | 5.96 | Parakeet TDT 0.6B v3 |
| Italian | 5.21 | 5.12 | 4.75 | Parakeet TDT 0.6B v3 |
| Portuguese | 6.53 | 6.44 | 6.27 | Parakeet TDT 0.6B v3 |
| Dutch | 9.21 | 9.45 | 7.93 | Parakeet TDT 0.6B v3 |
| Polish | 10.20 | 10.19 | 8.55 | Parakeet TDT 0.6B v3 |
| Russian | 8.67 | 9.01 | 7.21 | Parakeet TDT 0.6B v3 |
| Ukrainian | 7.89 | 8.05 | 7.27 | Parakeet TDT 0.6B v3 |
| Romanian | 11.88 | 11.89 | 13.13 | Phonon-2.5 |
| Czech | 10.40 | 11.41 | 11.93 | Phonon-2.5 |
| Slovak | 8.89 | 8.17 | 9.72 | Parakeet Redux |
| Bulgarian | 13.34 | 13.02 | 13.48 | Parakeet Redux |
| Croatian | 12.24 | 11.43 | 13.21 | Parakeet Redux |
| Swedish | 14.53 | 14.70 | 15.44 | Phonon-2.5 |
| Danish | 18.20 | 17.91 | 19.05 | Parakeet Redux |
| Finnish | 12.02 | 11.84 | 13.08 | Parakeet Redux |
| Estonian | 11.91 | 10.29 | 14.34 | Parakeet Redux |
| Latvian | 14.92 | 14.76 | 23.66 | Parakeet Redux |
| Lithuanian | 17.94 | 18.05 | 21.68 | Phonon-2.5 |
| Hungarian | 15.45 | 15.81 | 15.38 | Parakeet TDT 0.6B v3 |
| Slovenian | 18.53 | 18.88 | 24.70 | Phonon-2.5 |
| Maltese | 17.91 | 16.02 | 20.86 | Parakeet Redux |
| Greek | 20.83 | 33.25 | 36.11 | Phonon-2.5 |
| Average of 25 | 11.71 | 12.18 | 13.26 | Phonon-2.5 |

## Where each model is ahead

- Parakeet TDT 0.6B v3 is lower than Phonon-2.5 in: English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, Ukrainian, Hungarian. If your audio is mostly in these languages and a 2.5 GB model fits your hardware, it is the more accurate choice of the three.
- Parakeet Redux is lower than Phonon-2.5 in: Italian, Portuguese, Polish, Slovak, Bulgarian, Croatian, Danish, Finnish, Estonian, Latvian, Maltese.
- Phonon-2.5 is the lowest of the three in: Romanian, Czech, Swedish, Lithuanian, Slovenian, Greek.
- Greek is the largest gap: Phonon-2.5 20.83 %, Redux 33.25 %, Parakeet v3 36.11 %. Leave Greek out and Phonon-2.5 and Redux average 11.34 % and 11.31 % over the other 24, close to a tie; Parakeet v3 averages 12.31 %.

## Reading the table

Error rates vary far more by language than by model. Phonon-2.5 is under 10 % in 10 languages and at 15 % or more in 6, from 4.80 % in Spanish to 20.83 % in Greek. Languages with rich word forms, such as Finnish, Hungarian or the Baltic languages, tend to score higher on a word-level metric because one wrong ending costs a whole word.

FLEURS is read speech. Spontaneous speech, phone audio and meetings are harder, and these numbers are not a promise for that audio. Because our scoring keeps diacritics, these figures are not comparable with FLEURS numbers published elsewhere with other normalisers.

For scale on the same 25 sets and the same scoring: Whisper base, about the size of Phonon-2.5, averages 53.71 %.

## Questions

## Related

- [The FLEURS benchmark](https://www.fermionresearch.com/glossary/fleurs-benchmark/)
- [How Fermion Research measures speech models](https://www.fermionresearch.com/speech/benchmark-methodology/)
- [Phonon-2.5](https://www.fermionresearch.com/models/phonon-2-5/)
- [Phonon-2.5 vs Parakeet](https://www.fermionresearch.com/models/phonon-2-5/vs-parakeet/)
- [Multilingual speech recognition for European languages](https://www.fermionresearch.com/speech/multilingual-speech-recognition-european-languages/)
- [Introducing Phonon-2.5](https://www.fermionresearch.com/research/phonon-2-5/)

## Sources and method

- FLEURS numbers: Fermion Research measurement, published 2026-10-22, full FLEURS test set per language, one scoring code and normalisation for all three models, diacritics kept.
- Whisper base on the same 25 FLEURS sets: Phonon-2.5 release post, 2026-10-22.
- FLEURS dataset: huggingface.co/datasets/google/fleurs, read 2026-11-13.

Canonical: https://www.fermionresearch.com/speech/fleurs-results-25-languages/

Source: [https://www.fermionresearch.com/speech/fleurs-results-25-languages/](https://www.fermionresearch.com/speech/fleurs-results-25-languages/)
