FLEURS
Phi-4-multimodal 4.00% leads since March 2025 Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better.
FLEURS also supports speech translation, retrieval, and language identification, so check which task a score uses.
- Made by
- Released
- May 2022
- Kind
- Speech recognition and translation
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Last in a launch post
- September 2026
- Last model added
- October 2026
In the authors’ words
We introduce FLEURS, the Few-shot Learning Evaluation of Universal Representations of Speech benchmark. FLEURS is an n-way parallel speech dataset in 102 languages built on top of the machine translation FLoRes-101 benchmark, with approximately 12 hours of speech supervision per language. FLEURS can be used for a variety of speech tasks, including Automatic Speech Recognition (ASR), Speech Language Identification (Speech LangID), Translation and Retrieval. In this paper, we provide baselines for the tasks based on multilingual pre-trained models like mSLAM. The goal of FLEURS is to enable speech technology in more languages and catalyze research in low-resource speech understanding.
Phi-4-multimodal leads with 4.00% among models released by August 2026
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in August 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
-
Gemini 4 Carbon - GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 1 Phi-4-multimodal 4.00%
- 2
Qwen3-ASR-1.7B 4.90% - 3 Qwen3-Omni-30B-A3B 5.33%
- 4 Gemini 3.5 Transcribe 5.50%
- 5 Qwen3.5-Omni-Plus 6.55%
- 6 Qwen3.5-Omni-Flash 10.8%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Phi-4-multimodal | 4.00% | Mar 2025 | Lab model card | |
| 2 | Qwen3-ASR-1.7B | 4.90% | Jan 2026 | Lab model card | |
| 3 | Qwen3-Omni-30B-A3B | 5.33% | Sep 2025 | Lab model card | |
| 4 | Gemini 3.5 Transcribe | 5.50% | Aug 2026 | Lab model card | |
| 5 | Qwen3.5-Omni-Plus | 6.55% | Apr 2026 | Lab model card | |
| 6 | Qwen3.5-Omni-Flash | 10.8% | Mar 2026 | Lab model card |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Labs’ model cards
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Labs’ model cards
2 labs put it in a launch post in the last six months
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| Sep 18, 2026 | Qwen3.8-Omni-Flash | 9.3 | FLEURS-ASR Multilingual ASR (WER, 60 languages); FLEURS-S2TT BLEU 31.8 |
| Aug 26, 2026 | Gemini 3.5 Transcribe | 5.5 | FLEURS WER, streaming mode, top languages and locales |
| Jun 3, 2026 | Gemma 4 12B | 0.069 | FLEURS average WER, lower is better, excluding Chinese |
| Apr 28, 2026 |  Seed2.0 Lite (0428) | Fleurs (15 langs) (zh/en<->xx) speech translation | |
| Apr 2, 2026 | Gemma 4 E4B | 0.08 | FLEURS average WER, lower is better, excluding Chinese |
| Apr 2, 2026 | Gemma 4 E2B | 0.09 | FLEURS average WER, lower is better, excluding Chinese |
| Apr 2, 2026 | Gemma 4 31B IT | FLEURS (lower is better) (31B: no score) | |
| Mar 30, 2026 | Qwen3.5-Omni-Flash | 10.8 | Fleurs top60 ASR WER, Qwen3.5-Omni-Flash |
| Mar 30, 2026 | Qwen3.5-Omni-Plus | 6.55 | Fleurs(top60) ASR WER (S2TT xx to zh/en BLEU 32.8) |
| Jan 29, 2026 | Qwen3-ASR-1.7B | 4.9 | Fleurs multilingual WER, Qwen3-ASR-1.7B, 12 languages |
| Oct 31, 2025 |  LongCat-Flash-Omni | 3.99 | Fleurs (zh | en) 3.99 | 5.02 |
| Sep 22, 2025 | Qwen3-Omni-30B-A3B | 5.33 | Fleurs-avg (19 lang) ASR WER; Fleurs-en 2.72, Fleurs-zh 2.20 |
| Jul 15, 2025 |  Voxtral Small | FLEURS transcription, macro-average WER (chart) | |
| Apr 25, 2025 | Kimi-Audio-7B-Instruct | 2.69 | Fleurs (zh | en) 2.69 | 4.44 |
| Mar 27, 2025 | Qwen2.5-Omni-7B | 3.55 | Mean of Fleurs zh 3.0 / en 4.1 |
14 system cards report it, from 8 labs
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Oct 23, 2026 |  Nemotron 3.5 Bolt E4B technical report | 10.2 | FLEUR (WER/CER), lower is better, reasoning, 10.43 non-reasoning (Table 8) | Capabilities |
| Sep 22, 2026 | Qwen3.8-Omni-Flash technical report | 9.3 | FLEURS-ASR WER over 60 languages, lower is better (Table 4). FLEURS-S2TT BLEU 31.8 | Capabilities |
| Jul 24, 2026 | Gemma 4 E4B technical report | 0.075 | FLEURS average WER, lower is better (Table 7) | Capabilities |
| Jul 24, 2026 | Gemma 4 E2B technical report | 0.09 | FLEURS average WER, lower is better (Table 7) | Capabilities |
| Apr 17, 2026 | Qwen3.5-Omni-Plus technical report | 6.55 | Fleurs (top60) ASR WER, lower is better (Table 5). S2TT top59 BLEU: xx↔zh 30.2, xx↔en 35.4, xx↔zh/en 32.8 (Table 5). Per-language detail in Tables 13 to 15 | Capabilities |
| Feb 4, 2026 |  ERNIE 5.0 technical report | Fleurs-en 3.14, Fleurs-zh 0.83, WER, lower is better, no overall score (Table 7) | Capabilities | |
| Oct 31, 2025 | LongCat-Flash-Omni technical report | Fleurs zh 3.99, en 5.02, error rate, lower is better, no overall score (Table 9) | Capabilities | |
| Sep 22, 2025 | Qwen3-Omni-30B-A3B technical report | 5.33 | Fleurs-avg (19 lang) ASR WER, lower is better, Instruct (Table 6). Fleurs-en 2.72, Fleurs-zh 2.20. S2TT BLEU en2xx 37.50, xx2en 31.08, zh2xx 25.17, xx2zh 33.13 | Capabilities |
| Jul 17, 2025 | Voxtral Small technical report | FLEURS ASR WER per language with no printed macro-average, lower is better: ar 13.44, nl 4.94, en 3.35, fr 4.03, de 3.38, hi 7.69, it 2.62, pt 3.79, es 2.72 (Table 4, Figure 3 chart). FLEURS speech translation BLEU: en to de 47.0, es 39.9, fr 57.3, it 39.9, de to en 56.6, es to en 46.3, fr to en 54.2, it to en 46.8 (Table 7) | Capabilities | |
| Jul 7, 2025 | Gemini 2.5 Pro technical report | 6.66 | FLEURS, 53 languages, WER (lower is better) (Table 5) | Capabilities |
| Jul 7, 2025 | Gemini 2.5 Flash technical report | 9.95 | FLEURS, 53 languages, WER (lower is better) (Table 5) | Capabilities |
| Apr 25, 2025 | Kimi-Audio-7B-Instruct technical report | Fleurs zh 2.69, en 4.44, WER, lower is better, no overall score (Table 4) | Capabilities | |
| Mar 26, 2025 | Qwen2.5-Omni-7B technical report | Fleurs zh 3.0, en 4.1, WER, lower is better, no overall score (Table 2) | Capabilities | |
| Mar 3, 2025 | Phi-4-multimodal technical report | 4 | FLEURS ASR, WER average over 8 languages, lower is better (Tables 3 and 4). FLEURS AST BLEU X-EN 29.86 (CoT 32.35), EN-X 32.15 (CoT 33.56) (Table 5) | Capabilities |
More in Audio and video
- Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy. Qwen 3.8 Max 88.7%
- LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization. Gemini 4 Argon 91.7%
- MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy. Qwen 3.8 Max 82.4%
- MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy. Qwen 3.8 Max 90.8%
- VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules. Qwen3.5-Omni-Plus 93.1%
- TVBench In use Models answer multiple-choice questions on 10 temporal video tasks, such as action counting and ordering, that static frames cannot solve. Scored by accuracy. Qwen 3.8 Max 81.9%