Benchmark Almanac

PARADE Audio

gpt-4o-mini-transcribe transcription (GPT-4o + gpt-4o-mini-transcribe) Nov 2024 85.8% tops its leaderboard Qwen2.5-Omni-7B 63.4% leads as a model on its own
Not in model cards

Models answer questions about occupations and social status using audio clips. The evaluation tests whether voice cues bias their answers.

The release month follows the dataset repository history.

Released
March 2025
Kind
Bias, fairness and politics
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Propensity
Behavior
Bias
Language
English
In and out
Audio, Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
In views
Audio and speech
Strongest model tested
Gemini 2.5 Flash May 2025 #114 on the capability index
Last model added
May 2025
Scores refreshed
November 20th, 2026
17
Entries on AHELM’s leaderboard, 11 models we track

gpt-4o-mini-transcribe transcription (GPT-4o + gpt-4o-mini-transcribe) tops the leaderboard with 85.8% Qwen2.5-Omni-7B leads with 63.4% among models released by May 2025

Best score per entry from AHELM’s leaderboard, read on November 20th, 2026. Where a model ran on its own with several settings, we keep its best. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. The best model on its own is Qwen2.5-Omni-7B, at 63.4%. Agents built around a model, such as gpt-4o-mini-transcribe transcription (GPT-4o + gpt-4o-mini-transcribe), are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best.

The newest model with a score here came out in May 2025, 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-4o-mini-transcribe transcription (GPT-4o + gpt-4o-mini-transcribe) Nov 2024 85.8%
  2. 2 gpt-4o-transcribe transcription (GPT-4o + gpt-4o-transcribe) Nov 2024 85.8%
  3. 3 gpt-4o-mini-audio-preview-2024-12-17 85.7%
  4. 4 whisper-1 transcription (GPT-4o + whisper-1) Nov 2024 85.7%
  5. 5 gpt-4o-audio-preview-2024-10-01 84.7%
  6. 6 gpt-4o-audio-preview-2024-12-17 77.9%
  7. 7 Qwen2.5-Omni-7B 63.4%
  8. 8 Gemini 2.5 Flash May 2025 51.4%
  9. 9 Gemini 2.0 Flash Dec 2024 46.5%
  10. 10 Gemini 2.0 Flash Feb 2025 46.3%
  11. 11 Gemini 2.0 Flash-Lite 43.6%
  12. 12 Gemini 2.5 Pro May 2025 32.4%
  13. 13 Gemini 1.5 Flash Sep 2024 31.2%
  14. 14 Gemini 1.5 Flash May 2024 29.2%
  15. 15 Gemini 1.5 Pro May 2024 21.7%
  16. 16 Gemini 1.5 Pro Sep 2024 21.5%
  17. 17 Qwen2-Audio-7B-Instruct 20.9%

Source: AHELM’s leaderboard

#EntryScoreSettingDateSource
1 gpt-4o-mini-transcribe transcription (GPT-4o + gpt-4o-mini-transcribe) Nov 2024 85.8% Nov 2024 Official leaderboard
2 gpt-4o-transcribe transcription (GPT-4o + gpt-4o-transcribe) Nov 2024 85.8% Nov 2024 Official leaderboard
3 gpt-4o-mini-audio-preview-2024-12-17 85.7% Official leaderboard
4 whisper-1 transcription (GPT-4o + whisper-1) Nov 2024 85.7% Nov 2024 Official leaderboard
5 gpt-4o-audio-preview-2024-10-01 84.7% Official leaderboard
6 gpt-4o-audio-preview-2024-12-17 77.9% Official leaderboard
7 Qwen2.5-Omni-7B 63.4% Mar 2025 Official leaderboard
8 Gemini 2.5 Flash May 2025 51.4% May 2025 Official leaderboard
9 Gemini 2.0 Flash Dec 2024 46.5% Dec 2024 Official leaderboard
10 Gemini 2.0 Flash Feb 2025 46.3% Feb 2025 Official leaderboard
11 Gemini 2.0 Flash-Lite 43.6% Feb 2025 Official leaderboard
12 Gemini 2.5 Pro May 2025 32.4% May 2025 Official leaderboard
13 Gemini 1.5 Flash Sep 2024 31.2% Aug 2024 Official leaderboard
14 Gemini 1.5 Flash May 2024 29.2% May 2024 Official leaderboard
15 Gemini 1.5 Pro May 2024 21.7% May 2024 Official leaderboard
16 Gemini 1.5 Pro Sep 2024 21.5% Sep 2024 Official leaderboard
17 Qwen2-Audio-7B-Instruct 20.9% Aug 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Qwen2.5-Omni-7B 63.4% Mar 2025 Official leaderboard
2 Gemini 2.5 Flash May 2025 51.4% May 2025 Official leaderboard
3 Gemini 2.0 Flash Dec 2024 46.5% Dec 2024 Official leaderboard
4 Gemini 2.0 Flash Feb 2025 46.3% Feb 2025 Official leaderboard
5 Gemini 2.0 Flash-Lite 43.6% Feb 2025 Official leaderboard
6 Gemini 2.5 Pro May 2025 32.4% May 2025 Official leaderboard
7 Gemini 1.5 Flash Sep 2024 31.2% Aug 2024 Official leaderboard
8 Gemini 1.5 Flash May 2024 29.2% May 2024 Official leaderboard
9 Gemini 1.5 Pro May 2024 21.7% May 2024 Official leaderboard
10 Gemini 1.5 Pro Sep 2024 21.5% Sep 2024 Official leaderboard
11 Qwen2-Audio-7B-Instruct 20.9% Aug 2024 Official leaderboard

Rankings by part

We read 4 parts from AHELM’s leaderboard, each ranking models on one part of the test, such as occupation, female voice, status, female voice and occupation, male voice. Qwen2.5-Omni-7B leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.

occupation, female voice

11 models we track have a score here.

  1. 1 Qwen2.5-Omni-7B 44.4%
  2. 2 Gemini 2.5 Flash May 2025 34.4%
  3. 3 Gemini 2.0 Flash Dec 2024 28.7%
  4. 4 Gemini 2.0 Flash Feb 2025 28.7%
  5. 5 Gemini 2.0 Flash-Lite 28.5%
  6. 6 Gemini 2.5 Pro May 2025 15.2%
  7. 7 Qwen2-Audio-7B-Instruct ↑4 14.1%
  8. 8 Gemini 1.5 Pro May 2024 ↑1 8.94%
  9. 9 Gemini 1.5 Flash Sep 2024 ↓2 8.40%
  10. 10 Gemini 1.5 Pro Sep 2024 5.15%

2 models have held number one on AHELM’s leaderboard since its release in March 2025

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

Google DeepMind Qwen
AHELM leaderboard Not tested since May 2025

Source: AHELM’s 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 AHELM’s leaderboard Best so far
0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Qwen2.5-Omni-7B Gemini 2.0 Flash Gemini 1.5 Flash Gemini 1.5 Flash 0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Qwen2.5-Omni-7B Gemini 2.0 Flash

Source: AHELM’s 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 Safety and alignment

All 636 in Safety and alignment
  • FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
  • StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is. Inkling 98.6%
  • Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. Gemini 4 Carbon 0%
  • Covert misbehavior (Thinking Machines) In use Agentic tasks that check whether a model misbehaves and conceals it. Scored by the rate of covert misbehavior. Inkling 2 Mini 8.30%
  • Destructive actions (Thinking Machines) In use Agentic tasks that check whether a model takes destructive actions it was not authorized to take. Scored by the rate of destructive actions. Inkling 2 Small 4.70%
  • Misbehavior (Thinking Machines) In use Agentic tasks that check whether a model fabricates results, ignores instructions or hides what it did when a task is hard. Scored by the rate of misbehavior. Inkling 2 Mini 9.80%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 18 months ago, in May 2025, for Gemini 2.5 Flash. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (17 entries).
  • Pass Models scored 11 models we track, plus 6 other entries on its leaderboard.
  • Warning Newest score The newest came 18 months ago, in May 2025, for Gemini 2.5 Flash.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 2.5 Flash (#114).
  • Pass Headroom The best is 63.4%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on AHELM: 14 rows for models we track and 3 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