Benchmark Almanac

HearingBench

Gemini 4 Carbon 46% leads since October 2026
Not in model cards

Models identify sounds from spectrogram images without hearing the audio. The test covers everyday sounds, instruments, vocals, words and sentences.

The live leaderboard shows 16 model configurations on 100 clips, with Gemini 4 Carbon at 46 correct.

Made by
Adam Holter
Released
October 2026
Kind
Audio understanding and voice
Level
University
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Measures
Capability
Language
English
In and out
Images → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
In views
Vision
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last model added
November 2026
Scores refreshed
November 21st, 2026
16
Models scored on its own leaderboard

Gemini 4 Carbon leads with 46%

Best score per model from the benchmark's own leaderboard, read on November 21st, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

  1. 1 Gemini 4 Carbon medium 46%
  2. 2 Gemini 4 Neon high 45%
  3. 3 Gemini 4 Argon high 44%
  4. 4 Claude Opus 5.5 xhigh 43%
  5. 5 GPT-6.2 Astra max 43%
  6. 6 Claude Fable 5.2 xhigh 41%
  7. 7 GPT-6.2 Sol max 39%
  8. 8 Gemini 4 Hydrogen xhigh 38%
  9. 9 Gemini 3.8 Flash high 29%
  10. 10 GPT-6.1 Sol xhigh 26%
  11. 11 GPT-6 Astra xhigh 22%
  12. 12 Claude Sonnet 5.5 xhigh 20%
  13. 13 GPT-6.1 Terra high 18%
  14. 14 Grok 4.7 high 17%
  15. 15 GPT-6 Luna xhigh 9.00%
  16. 16 Claude Haiku 5.5 medium 9.00%

Source: Official HearingBench leaderboard

#ModelScoreSettingDateSource
1 Gemini 4 Carbon 46% medium, Antigravity Oct 2026 Official leaderboard
2 Gemini 4 Neon 45% high, Antigravity Nov 2026 Official leaderboard
3 Gemini 4 Argon 44% high, Antigravity Sep 2026 Official leaderboard
4 Claude Opus 5.5 43% xhigh, Claude Code Sep 2026 Official leaderboard
5 GPT-6.2 Astra 43% max, Codex Nov 2026 Official leaderboard
6 Claude Fable 5.2 41% xhigh, Claude Code Nov 2026 Official leaderboard
7 GPT-6.2 Sol 39% max, Codex Nov 2026 Official leaderboard
8 Gemini 4 Hydrogen 38% xhigh, Antigravity Nov 2026 Official leaderboard
9 Gemini 3.8 Flash 29% high, Antigravity Sep 2026 Official leaderboard
10 GPT-6.1 Sol 26% xhigh, Codex Sep 2026 Official leaderboard
11 GPT-6 Astra 22% xhigh, Codex Sep 2026 Official leaderboard
12 Claude Sonnet 5.5 20% xhigh, Claude Code Sep 2026 Official leaderboard
13 GPT-6.1 Terra 18% high, Codex Oct 2026 Official leaderboard
14 Grok 4.7 17% high, Cursor Sep 2026 Official leaderboard
15 GPT-6 Luna 9.00% xhigh, Codex Sep 2026 Official leaderboard
16 Claude Haiku 5.5 9.00% medium, Claude Code Oct 2026 Official leaderboard

More effort adds a median 12 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
Gemini 4 Carbon Antigravity
low medium
46%
Gemini 4 Argon Antigravity
low medium
44%
Claude Opus 5.5 Claude Code
low medium high xhigh
43%
GPT-6.2 Astra Codex
low medium xhigh max
43%
GPT-6.2 Sol Codex
low medium xhigh max
39%
Gemini 4 Hydrogen Antigravity
low medium
38%
Gemini 3.8 Flash Antigravity
low medium
29%
GPT-6.1 Sol Codex
low medium high
26%
GPT-6 Astra Codex
low high
22%
Claude Sonnet 5.5 Claude Code
low high xhigh
20%
GPT-6.1 Terra Codex
low medium
18%
Grok 4.7 Cursor
low medium
17%

Source: Official HearingBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Carbon medium Antigravity 46%
Gemini 4 Neon high Antigravity 45%
Gemini 4 Argon high Antigravity 44%
Gemini 4 Carbon high Antigravity 44%
Claude Opus 5.5 xhigh Claude Code 43%
Gemini 4 Argon medium Antigravity 43%
GPT-6.2 Astra max Codex 43%
Claude Fable 5.2 xhigh Claude Code 41%
Claude Opus 5.5 high Claude Code 39%
GPT-6.2 Sol max Codex 39%
Gemini 4 Hydrogen xhigh Antigravity 38%
GPT-6.2 Astra xhigh Codex 38%
Claude Opus 5.5 medium Claude Code 35%
Gemini 4 Hydrogen medium Antigravity 35%
Gemini 4 Hydrogen high Antigravity 34%
Claude Opus 5.5 low Claude Code 31%
Gemini 4 Carbon low Antigravity 31%
Gemini 4 Argon low Antigravity 30%
GPT-6.2 Sol xhigh Codex 30%
Gemini 3.8 Flash high Antigravity 29%
GPT-6.2 Astra medium Codex 28%
Gemini 3.8 Flash medium Antigravity 27%
GPT-6.2 Astra high Codex 27%
GPT-6.1 Sol xhigh Codex 26%
GPT-6.2 Sol high Codex 25%
GPT-6.2 Sol medium Codex 24%
GPT-6.1 Sol high Codex 23%
GPT-6 Astra xhigh Codex 22%
Gemini 4 Hydrogen low Antigravity 22%
GPT-6 Astra high Codex 21%
Claude Sonnet 5.5 xhigh Claude Code 20%
GPT-6.1 Sol low Codex 20%
GPT-6.2 Astra low Codex 20%
GPT-6.1 Terra high Codex 18%
GPT-6 Astra medium Codex 17%
Grok 4.7 high Cursor 17%
GPT-6.2 Sol low Codex 17%
GPT-6 Astra low Codex 16%
GPT-6.1 Sol medium Codex 16%
GPT-6.1 Terra medium Codex 16%
Gemini 3.8 Flash low Antigravity 15%
Grok 4.7 medium Cursor 15%
Claude Sonnet 5.5 high Claude Code 15%
GPT-6.1 Terra xhigh Codex 15%
Grok 4.7 xhigh Cursor 13%
Claude Sonnet 5.5 low Claude Code 11%
Grok 4.7 low Cursor 10%
Claude Sonnet 5.5 medium Claude Code 10%
GPT-6 Luna xhigh Codex 9.00%
Claude Haiku 5.5 medium Claude Code 9.00%
GPT-6 Luna high Codex 7.00%
Claude Haiku 5.5 low Claude Code 7.00%
Claude Haiku 5.5 high Claude Code 7.00%
Claude Haiku 5.5 xhigh Claude Code 7.00%
GPT-6.1 Terra low Codex 6.00%
GPT-6 Luna low Codex 3.00%
GPT-6 Luna medium Codex 2.00%

2 models have held number one on the official leaderboard since its release in October 2026

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

Google DeepMind
Own leaderboard 2 leaders

Source: Official HearingBench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Gemini 4 Carbon Gemini 3.8 Flash 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Gemini 4 Carbon Gemini 3.8 Flash

Source: Official HearingBench 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 11 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 5 pass
  • Pass Independent runs Scores from the official leaderboard (16 entries).
  • Pass Models scored 16 models we track.
  • Pass Newest score The newest is from November 2026, for Gemini 4 Neon.
  • Pass Frontier models 9 of the top 10 have a score, led by GPT-6.2 Astra (#1).
  • Pass Headroom The best is 46%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on hearing-bench.adamholter.chatgpt.site: 57 rows for models we track at the last read.
  • Pass Last read November 21st, 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