Benchmark Almanac

PAIRS (Parallel Images)

GPT-4o Nov 2024 97.2% leads since November 2024
Not in model cards

Models answer questions about people in paired images. The evaluation tests whether changing demographic characteristics changes judgments about the subjects.

Released
February 2024
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
Text, Images → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
In views
Vision
Strongest model tested
o3 #84 on the capability index
Last model added
April 2025
Scores refreshed
November 20th, 2026
43
Entries on VHELM’s leaderboard, 41 models we track

GPT-4o leads with 97.2% among models released by April 2025 GPT-4o leads with 97.2% among models released by April 2025

Best score per entry from VHELM’s leaderboard, read on November 20th, 2026. VHELM reports the percentage of responses that match the prefix of a correct reference after light processing. 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. This view shows only models we track, each at its best.

The newest model with a score here came out in April 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 Nov 2024 97.2%
  2. 2 GPT-4o Aug 2024 95.4%
  3. 3 GPT-4.1 nano 93%
  4. 4 Gemini 1.5 Pro Sep 2024 92.3%
  5. 5 o1 92.2%
  6. 6 GPT-4 Turbo Nov 2023 91.6%
  7. 7 Gemini 1.5 Pro May 2024, BLOCK_NONE safety 91.4%
  8. 8 GPT-4.5 89.8%
  9. 9 GPT-4.1 mini 89.2%
  10. 10 GPT-4o May 2024 87.3%
  11. 11 Gemini 2.0 Pro 87.2%
  12. 12 GPT-4 Turbo Apr 2024 86.1%
  13. 13 Gemini 2.0 Flash-Lite 84.8%
  14. 14 Gemini 2.5 Pro Mar 2025 83.8%
  15. 15 Qwen2-VL-72B-Instruct 82.8%
  16. 16 Qwen2.5-VL-32B 80.1%
  17. 17 GPT-4o mini 79.6%
  18. 18 Gemini 1.0 Pro Vision 78.8%
  19. 19 GPT-4.1 77.4%
  20. 20 o3 High reasoning effort 76.5%
  21. 21 Qwen2.5-VL-72B 75.6%
  22. 22 palmyra-vision-003 74%
  23. 23 Gemini 1.5 Flash May 2024, BLOCK_NONE safety 74%
  24. 24 Gemini 1.5 Flash Sep 2024 74%
  25. 25 o4-mini High reasoning effort 73.4%
  26. 26 Gemini 2.0 Flash Feb 2025 71.5%
  27. 27 Llama 4 Maverick 68.2%
  28. 28 Claude 3.5 Sonnet (October 2024) 65.4%
  29. 29 Qwen2.5-VL-7B 62.3%
  30. 30 Claude 3.5 Sonnet 61.4%
  31. 31 Claude 3 Opus 58.7%
  32. 32 Llama 4 Scout 55.4%
  33. 33 Qwen2-VL-7B-Instruct 48.8%
  34. 34 Idefics2-8B 48.7%
  35. 35 Pixtral 12B 46.6%
  36. 36 Qwen2.5-VL-3B 36.8%
  37. 37 PaliGemma-3B-mix-448 14%
  38. 38 Claude 3 Sonnet 13%
  39. 39 Claude 3 Haiku 8.03%
  40. 40 IDEFICS-9B-Instruct 8.00%
  41. 41 IDEFICS-80B-Instruct 7.33%
  42. 42 paligemma-3b-mix-224 0%
  43. 43 Qwen-VL-Chat 0%

Source: VHELM’s leaderboard

Source: VHELM’s leaderboard

#EntryScoreSettingDateSource
1 GPT-4o Nov 2024 97.2% Nov 2024 Official leaderboard
2 GPT-4o Aug 2024 95.4% Aug 2024 Official leaderboard
3 GPT-4.1 nano 93% Apr 2025 Official leaderboard
4 Gemini 1.5 Pro Sep 2024 92.3% Sep 2024 Official leaderboard
5 o1 92.2% Dec 2024 Official leaderboard
6 GPT-4 Turbo Nov 2023 91.6% Nov 2023 Official leaderboard
7 Gemini 1.5 Pro May 2024 91.4% BLOCK_NONE safety May 2024 Official leaderboard
8 GPT-4.5 89.8% Feb 2025 Official leaderboard
9 GPT-4.1 mini 89.2% Apr 2025 Official leaderboard
10 GPT-4o May 2024 87.3% May 2024 Official leaderboard
11 Gemini 2.0 Pro 87.2% Feb 2025 Official leaderboard
12 GPT-4 Turbo Apr 2024 86.1% Apr 2024 Official leaderboard
13 Gemini 2.0 Flash-Lite 84.8% Feb 2025 Official leaderboard
14 Gemini 2.5 Pro Mar 2025 83.8% Mar 2025 Official leaderboard
15 Qwen2-VL-72B-Instruct 82.8% Aug 2024 Official leaderboard
16 Qwen2.5-VL-32B 80.1% Mar 2025 Official leaderboard
17 GPT-4o mini 79.6% Jul 2024 Official leaderboard
18 Gemini 1.0 Pro Vision 78.8% Jan 2024 Official leaderboard
19 GPT-4.1 77.4% Apr 2025 Official leaderboard
20 o3 76.5% High reasoning effort Apr 2025 Official leaderboard
21 Qwen2.5-VL-72B 75.6% Jan 2025 Official leaderboard
22 palmyra-vision-003 74% Official leaderboard
23 Gemini 1.5 Flash May 2024 74% BLOCK_NONE safety May 2024 Official leaderboard
24 Gemini 1.5 Flash Sep 2024 74% Aug 2024 Official leaderboard
25 o4-mini 73.4% High reasoning effort Apr 2025 Official leaderboard
26 Gemini 2.0 Flash Feb 2025 71.5% Feb 2025 Official leaderboard
27 Llama 4 Maverick 68.2% Apr 2025 Official leaderboard
28 Claude 3.5 Sonnet (October 2024) 65.4% Oct 2024 Official leaderboard
29 Qwen2.5-VL-7B 62.3% Jan 2025 Official leaderboard
30 Claude 3.5 Sonnet 61.4% Jun 2024 Official leaderboard
31 Claude 3 Opus 58.7% Feb 2024 Official leaderboard
32 Llama 4 Scout 55.4% Apr 2025 Official leaderboard
33 Qwen2-VL-7B-Instruct 48.8% Aug 2024 Official leaderboard
34 Idefics2-8B 48.7% Apr 2024 Official leaderboard
35 Pixtral 12B 46.6% Sep 2024 Official leaderboard
36 Qwen2.5-VL-3B 36.8% Jan 2025 Official leaderboard
37 PaliGemma-3B-mix-448 14% May 2024 Official leaderboard
38 Claude 3 Sonnet 13% Feb 2024 Official leaderboard
39 Claude 3 Haiku 8.03% Mar 2024 Official leaderboard
40 IDEFICS-9B-Instruct 8.00% Jul 2023 Official leaderboard
41 IDEFICS-80B-Instruct 7.33% Jul 2023 Official leaderboard
42 paligemma-3b-mix-224 0% Official leaderboard
43 Qwen-VL-Chat 0% Aug 2023 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-4o Nov 2024 97.2% Nov 2024 Official leaderboard
2 GPT-4o Aug 2024 95.4% Aug 2024 Official leaderboard
3 GPT-4.1 nano 93% Apr 2025 Official leaderboard
4 Gemini 1.5 Pro Sep 2024 92.3% Sep 2024 Official leaderboard
5 o1 92.2% Dec 2024 Official leaderboard
6 GPT-4 Turbo Nov 2023 91.6% Nov 2023 Official leaderboard
7 Gemini 1.5 Pro May 2024 91.4% BLOCK_NONE safety May 2024 Official leaderboard
8 GPT-4.5 89.8% Feb 2025 Official leaderboard
9 GPT-4.1 mini 89.2% Apr 2025 Official leaderboard
10 GPT-4o May 2024 87.3% May 2024 Official leaderboard
11 Gemini 2.0 Pro 87.2% Feb 2025 Official leaderboard
12 GPT-4 Turbo Apr 2024 86.1% Apr 2024 Official leaderboard
13 Gemini 2.0 Flash-Lite 84.8% Feb 2025 Official leaderboard
14 Gemini 2.5 Pro Mar 2025 83.8% Mar 2025 Official leaderboard
15 Qwen2-VL-72B-Instruct 82.8% Aug 2024 Official leaderboard
16 Qwen2.5-VL-32B 80.1% Mar 2025 Official leaderboard
17 GPT-4o mini 79.6% Jul 2024 Official leaderboard
18 Gemini 1.0 Pro Vision 78.8% Jan 2024 Official leaderboard
19 GPT-4.1 77.4% Apr 2025 Official leaderboard
20 o3 76.5% High reasoning effort Apr 2025 Official leaderboard
21 Qwen2.5-VL-72B 75.6% Jan 2025 Official leaderboard
22 Gemini 1.5 Flash May 2024 74% BLOCK_NONE safety May 2024 Official leaderboard
23 Gemini 1.5 Flash Sep 2024 74% Aug 2024 Official leaderboard
24 o4-mini 73.4% High reasoning effort Apr 2025 Official leaderboard
25 Gemini 2.0 Flash Feb 2025 71.5% Feb 2025 Official leaderboard
26 Llama 4 Maverick 68.2% Apr 2025 Official leaderboard
27 Claude 3.5 Sonnet (October 2024) 65.4% Oct 2024 Official leaderboard
28 Qwen2.5-VL-7B 62.3% Jan 2025 Official leaderboard
29 Claude 3.5 Sonnet 61.4% Jun 2024 Official leaderboard
30 Claude 3 Opus 58.7% Feb 2024 Official leaderboard
31 Llama 4 Scout 55.4% Apr 2025 Official leaderboard
32 Qwen2-VL-7B-Instruct 48.8% Aug 2024 Official leaderboard
33 Idefics2-8B 48.7% Apr 2024 Official leaderboard
34 Pixtral 12B 46.6% Sep 2024 Official leaderboard
35 Qwen2.5-VL-3B 36.8% Jan 2025 Official leaderboard
36 PaliGemma-3B-mix-448 14% May 2024 Official leaderboard
37 Claude 3 Sonnet 13% Feb 2024 Official leaderboard
38 Claude 3 Haiku 8.03% Mar 2024 Official leaderboard
39 IDEFICS-9B-Instruct 8.00% Jul 2023 Official leaderboard
40 IDEFICS-80B-Instruct 7.33% Jul 2023 Official leaderboard
41 Qwen-VL-Chat 0% Aug 2023 Official leaderboard

3 models have held number one on VHELM’s leaderboard since its release in February 2024

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

OpenAI
VHELM leaderboard Passed 85% in Feb 2024, by GPT-4 Turbo
GPT-4o 97.2%

Source: VHELM’s leaderboard

GPT-4 Turbo on the official leaderboard passed 85% in November 2023

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
On VHELM’s leaderboard Best so far
0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) GPT-4o GPT-4 Turbo IDEFICS-9B-Instruct 0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) GPT-4o GPT-4 Turbo

Source: VHELM’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

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 (43 entries).
  • Pass Models scored 41 models we track, plus 2 other entries on its leaderboard.
  • Note Newest score The newest came 19 months ago, in April 2025, for o3. That’s expected for a saturated test.
  • Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is o3 (#84).
  • Note Headroom Saturated. The best is 97.2%, past 85% since November 2023.
Leaderboard import 3 pass
  • Pass Importer We read the board on VHELM: 46 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