speech to text benchmark
FLEURS
Few-shot Learning Evaluation of Universal Representations of Speech - a parallel speech dataset in 102 languages built on FLoRes-101 with approximately 12 hours of speech supervision per language for tasks including ASR, speech language identification, translation and retrieval. Scores are shown as speech recognition accuracy (1 - word error rate), so higher is better.
Updated Aug 7, 2026
Published models6
Registry coverage6
MetricScore
EvidenceB
FLEURS leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 95.9% | 100.0% | 6 | C | |
| 2 | GO | 93.6% | 80.0% | 6 | B | |
| 3 | GO | 93.3% | 60.0% | 6 | C | |
| 4 | GO | 93.1% | 40.0% | 6 | C | |
| 5 | GO | 90.4% | 20.0% | 6 | C | |
| 6 | GO | 86.4% | 0.0% | 6 | C |
Score distribution
Top published rows on the benchmark's original scale.
FLEURS highlights
The top published results on this benchmark's own scale.
What is FLEURS?
Definition and scoring fields from the benchmark registry.
Few-shot Learning Evaluation of Universal Representations of Speech - a parallel speech dataset in 102 languages built on FLoRes-101 with approximately 12 hours of speech supervision per language for tasks including ASR, speech language identification, translation and retrieval. Scores are shown as speech recognition accuracy (1 - word error rate), so higher is better.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- FLEURS
- Modality
- audio
- Primary category
- speech to text
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- fleurs|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about FLEURS.
Which model scores highest on FLEURS?
Qwen2.5-Omni-7B is currently ranked first with 95.9%.
What does FLEURS measure?
Few-shot Learning Evaluation of Universal Representations of Speech - a parallel speech dataset in 102 languages built on FLoRes-101 with approximately 12 hours of speech supervision per language for tasks including ASR, speech language identification, translation and retrieval. Scores are shown as speech recognition accuracy (1 - word error rate), so higher is better.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
6 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.