Benchmark Almanac

Audio Reasoning Challenge 2026

TalTech (challenge submission) 69.8% tops its leaderboard
Not in model cards

Models answer questions about speech, music, and environmental audio and explain their reasoning. Incorrect answers score zero. A model judge grades correct answers for reasoning quality, with the middle three of five evaluation runs averaged.

The challenge adds annotated reasoning and a new scoring protocol to MMAR. Its final stage uses 1,000 questions. Single-model and agent tracks have separate rankings.

Released
November 2025
Kind
Audio understanding and voice
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Audio → Text, Actions
Interaction
Agentic
Environment
APIs and tools
Items from
Another benchmark
Refresh
One fixed set
Run by
Anyone
Based on
MMAR
27
Entries on its own leaderboard, 0 models we track

TalTech (challenge submission) tops the leaderboard with 69.8%

Best score per entry from the benchmark's own leaderboard, read on November 20th, 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.

  1. 1 TalTech (challenge submission) agent track 69.8%
  2. 2 AISpeech (challenge submission) agent track 66.2%
  3. 3 AI^2 (challenge submission) agent track 66.1%
  4. 4 xianghe (challenge submission) single-model track 65.3%
  5. 5 AaronLi (challenge submission) agent track 64.6%
  6. 6 zsy20030814 (challenge submission) agent track 63%
  7. 7 tt (challenge submission) agent track 62.6%
  8. 8 tju-cca (challenge submission) single-model track 62.5%
  9. 9 sunsunsun (challenge submission) agent track 62.4%
  10. 10 TeleAI-NPU (challenge submission) single-model track 62.2%
  11. 11 alm-tf (challenge submission) agent track 61.9%
  12. 12 sujitnoronha (challenge submission) single-model track 60.6%
  13. 13 roysun2006 (challenge submission) agent track 60.3%
  14. 14 AudioMind (challenge submission) agent track 59.5%
  15. 15 (◍•ᴗ•◍) (challenge submission) single-model track 58.9%
  16. 16 jtjtjt (challenge submission) single-model track 58.5%
  17. 17 ABSTRACT (challenge submission) single-model track 57.8%
  18. 18 adasp (challenge submission) agent track 57.5%
  19. 19 zionhuang (challenge submission) single-model track 56%
  20. 20 artagent (challenge submission) agent track 54.2%
  21. 21 liusiqi (challenge submission) agent track 53.3%
  22. 22 dalietnguyen34 (challenge submission) agent track 53%
  23. 23 mimanchi (challenge submission) single-model track 47.8%
  24. 24 Aniket Tathe (challenge submission) single-model track 47.5%
  25. 25 erq111 (challenge submission) single-model track 38.9%
  26. 26 RobinLy (challenge submission) agent track 23.9%
  27. 27 bismarck91 (challenge submission) single-model track 19%

Source: Official Audio Reasoning Challenge 2026 leaderboard

#EntryScoreSettingSubmitted byDateSource
1 TalTech (challenge submission) 69.8% agent track TalTech (Tallinn University of Technology) Official leaderboard
2 AISpeech (challenge submission) 66.2% agent track AISpeech (AISpeech Co.Ltd, Shanghai Jiao Tong University) Official leaderboard
3 AI^2 (challenge submission) 66.1% agent track AI^2 (The Hong Kong University of Science and Technology (Guangzhou), Tencent AI Lab) Official leaderboard
4 xianghe (challenge submission) 65.3% single-model track xianghe (Tencent AI Lab, The Hong Kong University of Science and Technology (Guangzhou)) Official leaderboard
5 AaronLi (challenge submission) 64.6% agent track AaronLi (Hangzhou Yiwise Tech Co.Ltd) Official leaderboard
6 zsy20030814 (challenge submission) 63% agent track zsy20030814 (Tianjin University Tianjin Key Laboratory of Cognitive Computing and Application) Official leaderboard
7 tt (challenge submission) 62.6% agent track tt (Shanghai Jiao Tong University) Official leaderboard
8 tju-cca (challenge submission) 62.5% single-model track tju-cca (Tianjin University Tianjin Key Laboratory of Cognitive Computing and Application) Official leaderboard
9 sunsunsun (challenge submission) 62.4% agent track sunsunsun (ByteDance) Official leaderboard
10 TeleAI-NPU (challenge submission) 62.2% single-model track TeleAI-NPU (The Institute of Artificial Intelligence, China Telecom (TeleAI); Audio, Speech and Language Processing Lab, Northwestern Polytechnical University (NPU-ASLP)) Official leaderboard
11 alm-tf (challenge submission) 61.9% agent track alm-tf (Maum AI Inc.) Official leaderboard
12 sujitnoronha (challenge submission) 60.6% single-model track sujitnoronha (Independent Researcher) Official leaderboard
13 roysun2006 (challenge submission) 60.3% agent track roysun2006 (Ant Group) Official leaderboard
14 AudioMind (challenge submission) 59.5% agent track AudioMind (Southwestern University of Finance and Economics, Wuhan University, Xiaomi Inc.) Official leaderboard
15 (◍•ᴗ•◍) (challenge submission) 58.9% single-model track (◍•ᴗ•◍) (Bilibili Inc.) Official leaderboard
16 jtjtjt (challenge submission) 58.5% single-model track jtjtjt (China Mobile Jiutian Research) Official leaderboard
17 ABSTRACT (challenge submission) 57.8% single-model track ABSTRACT (National Yang Ming Chiao Tung University) Official leaderboard
18 adasp (challenge submission) 57.5% agent track adasp (Telecom Paris) Official leaderboard
19 zionhuang (challenge submission) 56% single-model track zionhuang (National Taiwan University) Official leaderboard
20 artagent (challenge submission) 54.2% agent track artagent (Guangzhou HKUST) Official leaderboard
21 liusiqi (challenge submission) 53.3% agent track liusiqi (ByteDance) Official leaderboard
22 dalietnguyen34 (challenge submission) 53% agent track dalietnguyen34 (VinUniversity) Official leaderboard
23 mimanchi (challenge submission) 47.8% single-model track mimanchi (Beijing Institute of Technology) Official leaderboard
24 Aniket Tathe (challenge submission) 47.5% single-model track Aniket Tathe (University of Illinois Urbana-Champaign, Carnegie Mellon University) Official leaderboard
25 erq111 (challenge submission) 38.9% single-model track erq111 (Tsinghua University) Official leaderboard
26 RobinLy (challenge submission) 23.9% agent track RobinLy (Ho Chi Minh City University of Science) Official leaderboard
27 bismarck91 (challenge submission) 19% single-model track bismarck91 (Johns Hopkins University) Official leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Audio and video

All 695 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%

Data audit

All 8 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
  • Pass Independent runs Scores from the official leaderboard (27 entries).
  • Pass Models scored 27 entries on its leaderboard, none a model we track.
  • Note Newest score No scores yet.
  • Note Frontier models No models scored yet.
  • Note Headroom No scores yet.
Leaderboard import 3 pass
  • Pass Importer We read the board on audio-reasoning-challenge.github.io: 27 entries at the last read, none a model we track.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check