Coval Speech-to-Text WER
reson8/realtime-global 1.41% tops its leaderboard 
Qwen3-ASR-1.7B 1.79% leads the models we track
Not in model cards Qwen3-ASR-1.7B 1.79% leads the models we track Speech models transcribe call-like audio. The score is mean word error percentage over the published pooled 24-hour window.
- Made by
- Coval
- Released
- April 2026
- Kind
- Speech recognition and translation
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Last model added
- August 2026
- Scores refreshed
- November 12th, 2026
29
Entries on its own leaderboard, 4 models we track
reson8/realtime-global tops the leaderboard with 1.41%
Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Qwen3-ASR-1.7B, at 1.79%.
The newest model with a score here came out in August 2026, and nobody has tested 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 reson8/realtime-global 24h pooled; mean WER (%) 1.41%
- 2 assemblyai/universal-3.6-pro 24h pooled; mean WER (%) 1.60%
- 3 assemblyai/universal-3.5-pro 24h pooled; mean WER (%) 1.64%
- 4 nari/qwen3-asr-fast 24h pooled; mean WER (%) 1.71%
- 5 Qwen3-ASR-1.7B 24h pooled; mean WER (%) 1.79%
- 6 reson8/realtime 24h pooled; mean WER (%) 1.86%
- 7 elevenlabs/scribe_v3_live 24h pooled; mean WER (%) 1.95%
- 8 Gemini 3.5 Transcribe Live 24h pooled; mean WER (%) 2.20%
- 9 google/chirp_3 24h pooled; mean WER (%) 2.46%
- 10 soniox/stt-rt-v6 24h pooled; mean WER (%) 2.58%
- 11 elevenlabs/scribe_v2_realtime 24h pooled; mean WER (%) 2.66%
- 12 Whisper large-v3 24h pooled; mean WER (%) 2.68%
- 13 saidwire/saidwire-en-2026-10-30 24h pooled; mean WER (%) 2.68%
- 14 speechmatics/linden-1 24h pooled; mean WER (%) 2.72%
- 15 google/chirp_2 24h pooled; mean WER (%) 2.78%
- 16
Voxtral Mini Transcribe Realtime (2602) 24h pooled; mean WER (%) 2.94% - 17 guava/daytona-stt 24h pooled; mean WER (%) 3.07%
- 18 openai/gpt-realtime-whisper 24h pooled; mean WER (%) 3.11%
- 19 smallest/pulse 24h pooled; mean WER (%) 3.31%
- 20 xai/grok-stt 24h pooled; mean WER (%) 3.40%
- 21 inworld/inworld-stt-1 24h pooled; mean WER (%) 3.53%
- 22 soniox/stt-rt-v5 24h pooled; mean WER (%) 3.74%
- 23 deepgram/nova-3 24h pooled; mean WER (%) 4.13%
- 24 gradium/default 24h pooled; mean WER (%) 4.44%
- 25 gladia/solaria-1 24h pooled; mean WER (%) 4.73%
- 26 deepgram/flux-general-en 24h pooled; mean WER (%) 4.84%
- 27 assemblyai/universal-streaming 24h pooled; mean WER (%) 5.03%
- 28 deepgram/nova-2 24h pooled; mean WER (%) 5.48%
- 29 deepgram/flux-general-multi 24h pooled; mean WER (%) 6.83%
Source: Official Coval Speech-to-Text WER leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | reson8/realtime-global | 1.41% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 2 | assemblyai/universal-3.6-pro | 1.60% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 3 | assemblyai/universal-3.5-pro | 1.64% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 4 | nari/qwen3-asr-fast | 1.71% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 5 | Qwen3-ASR-1.7B | 1.79% | 24h pooled; mean WER (%) | Jan 2026 | Official leaderboard |
| 6 | reson8/realtime | 1.86% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 7 | elevenlabs/scribe_v3_live | 1.95% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 8 | Gemini 3.5 Transcribe Live | 2.20% | 24h pooled; mean WER (%) | Aug 2026 | Official leaderboard |
| 9 | google/chirp_3 | 2.46% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 10 | soniox/stt-rt-v6 | 2.58% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 11 | elevenlabs/scribe_v2_realtime | 2.66% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 12 | Whisper large-v3 | 2.68% | 24h pooled; mean WER (%) | Nov 2023 | Official leaderboard |
| 13 | saidwire/saidwire-en-2026-10-30 | 2.68% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 14 | speechmatics/linden-1 | 2.72% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 15 | google/chirp_2 | 2.78% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 16 | Voxtral Mini Transcribe Realtime (2602) | 2.94% | 24h pooled; mean WER (%) | Feb 2026 | Official leaderboard |
| 17 | guava/daytona-stt | 3.07% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 18 | openai/gpt-realtime-whisper | 3.11% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 19 | smallest/pulse | 3.31% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 20 | xai/grok-stt | 3.40% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 21 | inworld/inworld-stt-1 | 3.53% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 22 | soniox/stt-rt-v5 | 3.74% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 23 | deepgram/nova-3 | 4.13% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 24 | gradium/default | 4.44% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 25 | gladia/solaria-1 | 4.73% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 26 | deepgram/flux-general-en | 4.84% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 27 | assemblyai/universal-streaming | 5.03% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 28 | deepgram/nova-2 | 5.48% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
| 29 | deepgram/flux-general-multi | 6.83% | 24h pooled; mean WER (%) | Nov 2026 | Official leaderboard |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Qwen
Source: Official Coval Speech-to-Text WER leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
On the benchmark’s own leaderboard
Best so far
Source: Official Coval Speech-to-Text WER leaderboard
No lab model card reports it yet
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.
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%
- FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better.
Phi-4-multimodal 4.00% - 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%