Audio Arena
Speech-to-speech models hold one long spoken conversation per scenario and call tools when needed, such as booking a dental appointment or ordering groceries. Six scenarios add up to 221 spoken turns with corrections, false memory traps, prompt injection and state tracking. An LLM judge grades each turn against a golden answer.
Turns use synthetic speech by default, with human recordings as an option. Claude is the default judge. The public board listed 14 speech-to-speech models on November 5th, 2026. Its repositories have under 10 GitHub stars, and no outside use was found.
- Made by
- Arcada Labs (DesignArena)
- Released
- February 2026
- Kind
- Speech recognition and translation
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- All public
- Last model added
- October 2026
- Scores refreshed
- November 12th, 2026
Grok Voice 3.0 leads with 85.7
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. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.
- 85.7 Grok Voice 3.0
- 85.4 GPT-Live-1
- 80.5 Grok Voice 2.0
- 77.8 ultravox-v0.7
- 77.5 GPT-Realtime-2.1
- 75.9 Grok Voice Think Fast 1.0
- 74.6 gemini-3.1-flash-live-preview
- 73.8 gemini-2.5-flash-native-audio-preview-12-2025
- 73.8 grok-realtime
- 73.7 GPT Realtime 1.5 (board label)
- 71.8 GPT-Realtime-2
- 58.5 amazon.nova-2-sonic-v1:0
- 26.9 glm-realtime-flash
- 1 Grok Voice 3.0 Human speech 85.7
- 2 GPT-Live-1 Artificial speech 85.4
- 3 Grok Voice 2.0 Artificial speech 80.5
- 4 ultravox-v0.7 Artificial speech 77.8
- 5 GPT-Realtime-2.1 Human speech 77.5
- 6 Grok Voice Think Fast 1.0 Human speech 75.9
- 7 gemini-3.1-flash-live-preview Artificial speech 74.6
- 8 gemini-2.5-flash-native-audio-preview-12-2025 Artificial speech 73.8
- 9 grok-realtime Artificial speech 73.8
- 10 GPT Realtime 1.5 (board label) Artificial speech 73.7
- 11 GPT-Realtime-2 Artificial speech 71.8
- 12 amazon.nova-2-sonic-v1:0 Artificial speech 58.5
- 13 glm-realtime-flash Artificial speech 26.9
Source: Official Audio Arena leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Grok Voice 3.0 | 85.7 | Human speech | Oct 2026 | Official leaderboard |
| 2 | GPT-Live-1 | 85.4 | Artificial speech | Jul 2026 | Official leaderboard |
| 3 | Grok Voice 2.0 | 80.5 | Artificial speech | Jul 2026 | Official leaderboard |
| 4 | ultravox-v0.7 | 77.8 | Artificial speech | Official leaderboard | |
| 5 | GPT-Realtime-2.1 | 77.5 | Human speech | Jul 2026 | Official leaderboard |
| 6 | Grok Voice Think Fast 1.0 | 75.9 | Human speech | Apr 2026 | Official leaderboard |
| 7 | gemini-3.1-flash-live-preview | 74.6 | Artificial speech | Official leaderboard | |
| 8 | gemini-2.5-flash-native-audio-preview-12-2025 | 73.8 | Artificial speech | Official leaderboard | |
| 9 | grok-realtime | 73.8 | Artificial speech | Official leaderboard | |
| 10 | GPT Realtime 1.5 (board label) | 73.7 | Artificial speech | Official leaderboard | |
| 11 | GPT-Realtime-2 | 71.8 | Artificial speech | May 2026 | Official leaderboard |
| 12 | amazon.nova-2-sonic-v1:0 | 58.5 | Artificial speech | Official leaderboard | |
| 13 | glm-realtime-flash | 26.9 | Artificial speech | Official leaderboard |
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: Official Audio Arena leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Grok Voice 3.0 | Human speech | 85.7 | |||
| GPT-Live-1 | Artificial speech | 85.4 | |||
| GPT-Live-1 | Human speech | 85.1 | |||
| Grok Voice 3.0 | Artificial speech | 83.0 | |||
| Grok Voice 2.0 | Artificial speech | 80.5 | |||
| Grok Voice 2.0 | Human speech | 77.7 | |||
| GPT-Realtime-2.1 | Human speech | 77.5 | |||
| Grok Voice Think Fast 1.0 | Human speech | 75.9 | |||
| Grok Voice Think Fast 1.0 | Artificial speech | 75.9 | |||
| GPT-Realtime-2.1 | Artificial speech | 71.9 | |||
| GPT-Realtime-2 | Artificial speech | 71.8 | |||
| GPT-Realtime-2 | Human speech | 67.6 |
4 models have held number one on the official leaderboard since its release in February 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official Audio Arena leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Audio Arena 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%