Benchmark Almanac

PianoBench

GPT-6 Astra 93.2% leads since November 2026
Not in model cards

Models answer questions about piano performances from recordings or symbolic transcriptions. The authors compare accuracy with and without access to the music.

Released
November 2026
Kind
Audio understanding and voice
Level
University
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Subject
Arts
Language
English
In and out
Audio, Text → Text
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
In views
Audio and speech
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026
18
Entries in its paper, 16 models we track

GPT-6 Astra leads with 93.2% among models released by September 2026 GPT-6 Astra leads with 93.2% among models released by September 2026

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. 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. This view shows only models we track, each at its best.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-6 Astra Zero-shot, MIDI-text, high 93.2%
  2. 2 Gemini 3.7 Flash Zero-shot, MIDI-text, medium 90%
  3. 3 Gemini 3.1 Pro Zero-shot, MIDI-text, medium 89.2%
  4. 4 Gemini 3.8 Flash Zero-shot, MIDI-text, medium 88.8%
  5. 5 GPT-5.6 Sol Zero-shot, MIDI-text, xhigh 88%
  6. 6 Muse Spark 1.2 Zero-shot, MIDI-text, medium 86.8%
  7. 7 GPT-5.6 Terra Zero-shot, MIDI-text, high 85.2%
  8. 8 Qwen3.5-9B Zero-shot, MIDI-text, thinking 70%
  9. 9 Inkling-Small Zero-shot, MIDI-text, medium 68.8%
  10. 10 Inkling Zero-shot, MIDI-text, medium 66.4%
  11. 11 Qwen3.5-Omni-Plus Zero-shot, Piano stem 63.2%
  12. 12 Gemma 4 12B Zero-shot, MIDI-text, thinking 58.4%
  13. 13 Qwen3-Omni-30B-A3B Zero-shot, Piano synth, non-thinking 50.8%
  14. 14 Nemotron Audex 30B-A3B Zero-shot, MIDI-text 48.8%
  15. 15 Qwen3-Omni-30B-A3B-Thinking Zero-shot, Piano synth, thinking 48.4%
  16. 16 Nemotron 3 Nano Omni 30B-A3B Zero-shot, Piano stem 48.4%
  17. 17 Music Flamingo Zero-shot, Piano synth 45.6%
  18. 18 Audio Flamingo 3 Zero-shot, Piano synth 38.8%

Source: The PianoBench paper

#EntryScoreSettingDateSource
1 GPT-6 Astra 93.2% Zero-shot, MIDI-text, high Sep 2026 Official leaderboard
2 Gemini 3.7 Flash 90% Zero-shot, MIDI-text, medium Aug 2026 Official leaderboard
3 Gemini 3.1 Pro 89.2% Zero-shot, MIDI-text, medium Feb 2026 Official leaderboard
4 Gemini 3.8 Flash 88.8% Zero-shot, MIDI-text, medium Sep 2026 Official leaderboard
5 GPT-5.6 Sol 88% Zero-shot, MIDI-text, xhigh Jul 2026 Official leaderboard
6 Muse Spark 1.2 86.8% Zero-shot, MIDI-text, medium Aug 2026 Official leaderboard
7 GPT-5.6 Terra 85.2% Zero-shot, MIDI-text, high Jul 2026 Official leaderboard
8 Qwen3.5-9B 70% Zero-shot, MIDI-text, thinking Feb 2026 Official leaderboard
9 Inkling-Small 68.8% Zero-shot, MIDI-text, medium Jul 2026 Official leaderboard
10 Inkling 66.4% Zero-shot, MIDI-text, medium Jul 2026 Official leaderboard
11 Qwen3.5-Omni-Plus 63.2% Zero-shot, Piano stem Mar 2026 Official leaderboard
12 Gemma 4 12B 58.4% Zero-shot, MIDI-text, thinking Jun 2026 Official leaderboard
13 Qwen3-Omni-30B-A3B 50.8% Zero-shot, Piano synth, non-thinking Sep 2025 Official leaderboard
14 Nemotron Audex 30B-A3B 48.8% Zero-shot, MIDI-text Official leaderboard
15 Qwen3-Omni-30B-A3B-Thinking 48.4% Zero-shot, Piano synth, thinking Sep 2025 Official leaderboard
16 Nemotron 3 Nano Omni 30B-A3B 48.4% Zero-shot, Piano stem Apr 2026 Official leaderboard
17 Music Flamingo 45.6% Zero-shot, Piano synth Official leaderboard
18 Audio Flamingo 3 38.8% Zero-shot, Piano synth Jul 2025 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6 Astra 93.2% Zero-shot, MIDI-text, high Sep 2026 Official leaderboard
2 Gemini 3.7 Flash 90% Zero-shot, MIDI-text, medium Aug 2026 Official leaderboard
3 Gemini 3.1 Pro 89.2% Zero-shot, MIDI-text, medium Feb 2026 Official leaderboard
4 Gemini 3.8 Flash 88.8% Zero-shot, MIDI-text, medium Sep 2026 Official leaderboard
5 GPT-5.6 Sol 88% Zero-shot, MIDI-text, xhigh Jul 2026 Official leaderboard
6 Muse Spark 1.2 86.8% Zero-shot, MIDI-text, medium Aug 2026 Official leaderboard
7 GPT-5.6 Terra 85.2% Zero-shot, MIDI-text, high Jul 2026 Official leaderboard
8 Qwen3.5-9B 70% Zero-shot, MIDI-text, thinking Feb 2026 Official leaderboard
9 Inkling-Small 68.8% Zero-shot, MIDI-text, medium Jul 2026 Official leaderboard
10 Inkling 66.4% Zero-shot, MIDI-text, medium Jul 2026 Official leaderboard
11 Qwen3.5-Omni-Plus 63.2% Zero-shot, Piano stem Mar 2026 Official leaderboard
12 Gemma 4 12B 58.4% Zero-shot, MIDI-text, thinking Jun 2026 Official leaderboard
13 Qwen3-Omni-30B-A3B 50.8% Zero-shot, Piano synth, non-thinking Sep 2025 Official leaderboard
14 Qwen3-Omni-30B-A3B-Thinking 48.4% Zero-shot, Piano synth, thinking Sep 2025 Official leaderboard
15 Nemotron 3 Nano Omni 30B-A3B 48.4% Zero-shot, Piano stem Apr 2026 Official leaderboard
16 Audio Flamingo 3 38.8% Zero-shot, Piano synth Jul 2025 Official leaderboard

More effort adds a median 28 points

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.

Lowest effort to highest, light to dark
GPT-6 Astra
Zero-shot, … Zero-shot, …
93.2%
Gemini 3.7 Flash
Zero-shot, … Zero-shot, … Zero-shot, …
90%
Gemini 3.1 Pro
Zero-shot, … Zero-shot, … Zero-shot, …
89.2%
Gemini 3.8 Flash
Zero-shot, … Zero-shot, … Zero-shot, …
88.8%
GPT-5.6 Sol
Zero-shot, … Zero-shot, … Zero-shot, … Zero-shot, … Zero-shot, … Zero-shot, … Zero-shot, … Zero-shot, …
88%
Muse Spark 1.2
Zero-shot, … Zero-shot, … Zero-shot, …
86.8%
GPT-5.6 Terra
Zero-shot, … Zero-shot, …
85.2%
Qwen3.5-9B
Zero-shot, … Zero-shot, … Zero-shot, …
70%
Inkling-Small
Zero-shot, … Zero-shot, …
68.8%
Inkling
Zero-shot, … Zero-shot, …
66.4%
Qwen3.5-Omni-Plus
Zero-shot, … Zero-shot, … Zero-shot, …
63.2%
Gemma 4 12B
Zero-shot, … Zero-shot, … Zero-shot, …
58.4%

Source: The PianoBench paper

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra Zero-shot, MIDI-text, high 93.2%
Gemini 3.7 Flash Zero-shot, MIDI-text, medium 90%
Gemini 3.1 Pro Zero-shot, MIDI-text, medium 89.2%
Gemini 3.8 Flash Zero-shot, MIDI-text, medium 88.8%
GPT-5.6 Sol Zero-shot, MIDI-text, xhigh 88%
GPT-5.6 Sol Zero-shot, MIDI-text, high 87.2%
Muse Spark 1.2 Zero-shot, MIDI-text, medium 86.8%
GPT-5.6 Sol Zero-shot, MIDI-text, medium 85.6%
GPT-5.6 Terra Zero-shot, MIDI-text, high 85.2%
Gemini 3.1 Pro Zero-shot, Piano stem, medium 80.4%
Gemini 3.1 Pro Zero-shot, Piano synth, medium 78.4%
GPT-5.6 Sol Zero-shot, MIDI-text, none 70.4%
Qwen3.5-9B Zero-shot, MIDI-text, thinking 70%
Inkling-Small Zero-shot, MIDI-text, medium 68.8%
Gemini 3.7 Flash Zero-shot, Piano stem, medium 66.8%
Inkling Zero-shot, MIDI-text, medium 66.4%
Gemini 3.8 Flash Zero-shot, Piano synth, medium 66.4%
Gemini 3.8 Flash Zero-shot, Piano stem, medium 64%
Gemini 3.1 Pro Zero-shot, No music, medium 63.6%
Gemini 3.7 Flash Zero-shot, Piano synth, medium 63.6%
Qwen3.5-Omni-Plus Zero-shot, Piano stem 63.2%
Qwen3.5-Omni-Plus Zero-shot, Piano synth 62.4%
Gemini 3.7 Flash Zero-shot, No music, medium 62%
Gemini 3.8 Flash Zero-shot, No music, medium 62%
GPT-5.6 Sol Zero-shot, No music, xhigh 58.8%
Gemma 4 12B Zero-shot, MIDI-text, thinking 58.4%
Qwen3.5-Omni-Plus Zero-shot, MIDI-text 57.2%
Muse Spark 1.2 Zero-shot, Piano synth, medium 56.8%
GPT-6 Astra Zero-shot, No music, high 56.8%
GPT-5.6 Sol Zero-shot, No music, high 56.4%
GPT-5.6 Sol Zero-shot, No music, medium 56%
Muse Spark 1.2 Zero-shot, Piano stem, medium 55.6%
Inkling-Small Zero-shot, Piano synth, medium 52.8%
GPT-5.6 Sol Zero-shot, No music, none 52.4%
Qwen3.5-Omni-Plus Zero-shot, No music 52%
Gemma 4 12B Zero-shot, Piano stem, non-thinking 51.6%
Inkling-Small Zero-shot, No music, medium 51.6%
Gemma 4 12B Zero-shot, Piano synth, non-thinking 51.2%
Qwen3-Omni-30B-A3B Zero-shot, Piano synth, non-thinking 50.8%
Inkling-Small Zero-shot, Piano stem, medium 50.4%
Qwen3.5-9B Zero-shot, MIDI-text, non-thinking 50%
Qwen3-Omni-30B-A3B Zero-shot, Piano stem, non-thinking 49.2%
Muse Spark 1.2 Zero-shot, No music, medium 48.8%
Qwen3-Omni-30B-A3B-Thinking Zero-shot, Piano synth, thinking 48.4%
Nemotron 3 Nano Omni 30B-A3B Zero-shot, Piano stem 48.4%
Qwen3-Omni-30B-A3B-Thinking Zero-shot, MIDI-text, thinking 47.6%
Inkling Zero-shot, Piano synth, medium 47.6%
Qwen3-Omni-30B-A3B Zero-shot, MIDI-text, non-thinking 47.2%
Nemotron 3 Nano Omni 30B-A3B Zero-shot, Piano synth 47.2%
GPT-5.6 Terra Zero-shot, No music, high 47.2%
Inkling Zero-shot, No music, medium 47.2%
Qwen3-Omni-30B-A3B-Thinking Zero-shot, Piano stem, thinking 46.8%
Inkling Zero-shot, Piano stem, medium 46.8%
Qwen3.5-9B Zero-shot, No music, non-thinking 46.4%
Nemotron 3 Nano Omni 30B-A3B Zero-shot, No music 46.4%
Gemma 4 12B Zero-shot, Piano synth, thinking 46.4%
Nemotron 3 Nano Omni 30B-A3B Zero-shot, MIDI-text 45.6%
Gemma 4 12B Zero-shot, Piano stem, thinking 45.2%
Qwen3-Omni-30B-A3B Zero-shot, No music, non-thinking 44.8%
Gemma 4 12B Zero-shot, No music, thinking 44.8%
Qwen3.5-9B Zero-shot, No music, thinking 43.6%
Qwen3-Omni-30B-A3B-Thinking Zero-shot, No music, thinking 42.4%
Audio Flamingo 3 Zero-shot, Piano synth 38.8%
Audio Flamingo 3 Zero-shot, Piano stem 37.2%

More effort adds a median 28 points

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.

Lowest effort to highest, light to dark
GPT-6 Astra
Zero-shot, … Zero-shot, …
93.2%
Gemini 3.7 Flash
Zero-shot, … Zero-shot, … Zero-shot, …
90%
Gemini 3.1 Pro
Zero-shot, … Zero-shot, … Zero-shot, …
89.2%
Gemini 3.8 Flash
Zero-shot, … Zero-shot, … Zero-shot, …
88.8%
GPT-5.6 Sol
Zero-shot, … Zero-shot, … Zero-shot, … Zero-shot, … Zero-shot, … Zero-shot, … Zero-shot, … Zero-shot, …
88%
Muse Spark 1.2
Zero-shot, … Zero-shot, … Zero-shot, …
86.8%
GPT-5.6 Terra
Zero-shot, … Zero-shot, …
85.2%
Qwen3.5-9B
Zero-shot, … Zero-shot, … Zero-shot, …
70%
Inkling-Small
Zero-shot, … Zero-shot, …
68.8%
Inkling
Zero-shot, … Zero-shot, …
66.4%
Qwen3.5-Omni-Plus
Zero-shot, … Zero-shot, … Zero-shot, …
63.2%
Gemma 4 12B
Zero-shot, … Zero-shot, … Zero-shot, …
58.4%

Source: The PianoBench paper

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra Zero-shot, MIDI-text, high 93.2%
Gemini 3.7 Flash Zero-shot, MIDI-text, medium 90%
Gemini 3.1 Pro Zero-shot, MIDI-text, medium 89.2%
Gemini 3.8 Flash Zero-shot, MIDI-text, medium 88.8%
GPT-5.6 Sol Zero-shot, MIDI-text, xhigh 88%
GPT-5.6 Sol Zero-shot, MIDI-text, high 87.2%
Muse Spark 1.2 Zero-shot, MIDI-text, medium 86.8%
GPT-5.6 Sol Zero-shot, MIDI-text, medium 85.6%
GPT-5.6 Terra Zero-shot, MIDI-text, high 85.2%
Gemini 3.1 Pro Zero-shot, Piano stem, medium 80.4%
Gemini 3.1 Pro Zero-shot, Piano synth, medium 78.4%
GPT-5.6 Sol Zero-shot, MIDI-text, none 70.4%
Qwen3.5-9B Zero-shot, MIDI-text, thinking 70%
Inkling-Small Zero-shot, MIDI-text, medium 68.8%
Gemini 3.7 Flash Zero-shot, Piano stem, medium 66.8%
Inkling Zero-shot, MIDI-text, medium 66.4%
Gemini 3.8 Flash Zero-shot, Piano synth, medium 66.4%
Gemini 3.8 Flash Zero-shot, Piano stem, medium 64%
Gemini 3.1 Pro Zero-shot, No music, medium 63.6%
Gemini 3.7 Flash Zero-shot, Piano synth, medium 63.6%
Qwen3.5-Omni-Plus Zero-shot, Piano stem 63.2%
Qwen3.5-Omni-Plus Zero-shot, Piano synth 62.4%
Gemini 3.7 Flash Zero-shot, No music, medium 62%
Gemini 3.8 Flash Zero-shot, No music, medium 62%
GPT-5.6 Sol Zero-shot, No music, xhigh 58.8%
Gemma 4 12B Zero-shot, MIDI-text, thinking 58.4%
Qwen3.5-Omni-Plus Zero-shot, MIDI-text 57.2%
Muse Spark 1.2 Zero-shot, Piano synth, medium 56.8%
GPT-6 Astra Zero-shot, No music, high 56.8%
GPT-5.6 Sol Zero-shot, No music, high 56.4%
GPT-5.6 Sol Zero-shot, No music, medium 56%
Muse Spark 1.2 Zero-shot, Piano stem, medium 55.6%
Inkling-Small Zero-shot, Piano synth, medium 52.8%
GPT-5.6 Sol Zero-shot, No music, none 52.4%
Qwen3.5-Omni-Plus Zero-shot, No music 52%
Gemma 4 12B Zero-shot, Piano stem, non-thinking 51.6%
Inkling-Small Zero-shot, No music, medium 51.6%
Gemma 4 12B Zero-shot, Piano synth, non-thinking 51.2%
Qwen3-Omni-30B-A3B Zero-shot, Piano synth, non-thinking 50.8%
Inkling-Small Zero-shot, Piano stem, medium 50.4%
Qwen3.5-9B Zero-shot, MIDI-text, non-thinking 50%
Qwen3-Omni-30B-A3B Zero-shot, Piano stem, non-thinking 49.2%
Muse Spark 1.2 Zero-shot, No music, medium 48.8%
Qwen3-Omni-30B-A3B-Thinking Zero-shot, Piano synth, thinking 48.4%
Nemotron 3 Nano Omni 30B-A3B Zero-shot, Piano stem 48.4%
Qwen3-Omni-30B-A3B-Thinking Zero-shot, MIDI-text, thinking 47.6%
Inkling Zero-shot, Piano synth, medium 47.6%
Qwen3-Omni-30B-A3B Zero-shot, MIDI-text, non-thinking 47.2%
Nemotron 3 Nano Omni 30B-A3B Zero-shot, Piano synth 47.2%
GPT-5.6 Terra Zero-shot, No music, high 47.2%
Inkling Zero-shot, No music, medium 47.2%
Qwen3-Omni-30B-A3B-Thinking Zero-shot, Piano stem, thinking 46.8%
Inkling Zero-shot, Piano stem, medium 46.8%
Qwen3.5-9B Zero-shot, No music, non-thinking 46.4%
Nemotron 3 Nano Omni 30B-A3B Zero-shot, No music 46.4%
Gemma 4 12B Zero-shot, Piano synth, thinking 46.4%
Nemotron 3 Nano Omni 30B-A3B Zero-shot, MIDI-text 45.6%
Gemma 4 12B Zero-shot, Piano stem, thinking 45.2%
Qwen3-Omni-30B-A3B Zero-shot, No music, non-thinking 44.8%
Gemma 4 12B Zero-shot, No music, thinking 44.8%
Qwen3.5-9B Zero-shot, No music, thinking 43.6%
Qwen3-Omni-30B-A3B-Thinking Zero-shot, No music, thinking 42.4%
Audio Flamingo 3 Zero-shot, Piano synth 38.8%
Audio Flamingo 3 Zero-shot, Piano stem 37.2%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Benchmark paper Passed 85% in Nov 2026, by GPT-6 Astra
GPT-6 Astra 93.2%

Source: The PianoBench paper

Gemini 3.1 Pro on the official leaderboard passed 85% in February 2026

Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-6 Astra Gemini 3.1 Pro Qwen3-Omni-30B-A3B Audio Flamingo 3 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-6 Astra Qwen3-Omni-30B-A3B

Source: Official PianoBench 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 9 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (18 entries).
  • Pass Models scored 16 models we track, plus 2 other entries on its leaderboard.
  • Pass Newest score The newest is from September 2026, for GPT-6 Astra.
  • Note Frontier models 1 of the top 10 has a score, which is fine for a retired test. The strongest tested is GPT-6 Astra (#4).
  • Note Headroom Saturated. The best is 93.2%, past 85% since February 2026.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 64 rows for models we track and 2 other entries at the last read.
  • 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