Big Bench Audio
Models answer 1,000 spoken questions from four Big Bench Hard tasks: formal fallacies, navigate, object counting and web of lies. Built to test reasoning over audio input.
Audio was generated with 23 synthetic voices. Used for speech-to-speech and audio-to-text models. MIT license.
- Level
- School Students, from grade school to high school
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Short answer A number, word or short string matched to a key.
Grok Voice 3.0 leads with 98.5%
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.
- 1 Grok Voice 3.0 98.5%
- 2 Grok Voice 2.0 97.2%
- 3 GPT-Realtime-2 96.6%
- 4 MiMo-V2-Omni 94%
- 5 gpt-realtime 82.8%
- 6 MiMo-Audio-7B-Instruct 72.9%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Grok Voice 3.0 | 98.5% | Oct 2026 | Model card | |
| 2 | Grok Voice 2.0 | 97.2% | Jul 2026 | Model card | |
| 3 | GPT-Realtime-2 | 96.6% | May 2026 | Model card | |
| 4 | MiMo-V2-Omni | 94% | Mar 2026 | Model card | |
| 5 | gpt-realtime | 82.8% | Aug 2025 | Model card | |
| 6 | MiMo-Audio-7B-Instruct | 72.9% | Sep 2025 | Model card |
5 models have held number one since its release in Dec 2024
Lab claims count from the day the card came out, independent runs from the model’s release.
The best score passed 85% in Mar 2026
All the scores we have, one dot each, so a model tested at several settings appears more than once.
One lab 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 |
|---|---|---|---|
| 14 Oct 2026 | Grok Voice 3.0 | 98.5 | Speech Reasoning Big Bench Audio (per Artificial Analysis) |
| 29 Jul 2026 | Grok Voice 2.0 | 97.2 | Speech Reasoning Big Bench Audio (per Artificial Analysis) |
| 7 May 2026 | GPT-Realtime-2 | 96.6 | Big Bench Audio (Intelligence), gpt-realtime-2 (high) |
| 18 Mar 2026 | MiMo-V2-Omni | 94 | BigBench Audio |
| 18 Sep 2025 | MiMo-Audio-7B-Instruct | 72.9 | Best of Big Bench Audio S2T 72.9 / S2S 60.2 |
| 28 Aug 2025 | gpt-realtime | 82.8 | Big Bench Audio (Intelligence) |
One system card reports it
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 |
|---|---|---|---|---|
| 18 Sep 2025 | MiMo-Audio-7B-Instruct technical report | 72.9 | Big Bench Audio S2T 72.90, S2S 60.20, GPT-4o-mini judge (Table 8) | 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%