VisionArena-Bench
Models answer visual questions from Arena users. A vision language model judges their answers to approximate human preference rankings.
The offline benchmark contains 500 prompts.
- Made by
- Arena
- Released
- December 2024
- Kind
- General visual questions
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Preference
- Language
- English
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- Based on
- Vision Arena
- In views
- Vision
- Strongest model tested
-
Gemini 1.5 Pro Sep 2024 #157 on the capability index
- Last model added
- October 2024
- Scores refreshed
- November 20th, 2026
Gemini 1.5 Pro leads with 88 among models released by October 2024
Best score per model 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.
The newest model with a score here came out in October 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1
Gemini 1.5 Pro Sep 2024, Baseline: claude-3-sonnet-20240229 88
- 2 GPT-4o May 2024, Baseline: claude-3-sonnet-20240229 87
- 3 Claude 3.5 Sonnet Baseline: claude-3-sonnet-20240229 86
- 4
Gemini 1.5 Flash Sep 2024, Baseline: claude-3-sonnet-20240229 84
- 5 GPT-4 Turbo Apr 2024, Baseline: claude-3-sonnet-20240229 81
- 6
Gemini 1.5 Pro May 2024, Baseline: claude-3-sonnet-20240229 75
- 7 GPT-4o mini Baseline: claude-3-sonnet-20240229 73
- 8
Gemini 1.5 Flash 8B Baseline: claude-3-sonnet-20240229 65
- 9
Gemini 1.5 Flash May 2024, Baseline: claude-3-sonnet-20240229 58
- 10
InternVL2-26B Baseline: claude-3-sonnet-20240229 54
- 11 Claude 3 Opus Baseline: claude-3-sonnet-20240229 52
- 12 Claude 3 Sonnet Baseline: claude-3-sonnet-20240229 50
- 13 Reka Core Baseline: claude-3-sonnet-20240229 38
- 14
Llama 3.2 11B Baseline: claude-3-sonnet-20240229 33
- 15 Claude 3 Haiku Baseline: claude-3-sonnet-20240229 31
- 16
InternVL2-4B Baseline: claude-3-sonnet-20240229 20
Source: The VisionArena-Bench paper
| # | Model | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 88 ±1.1 | Baseline: claude-3-sonnet-20240229 | Sep 2024 | Official leaderboard |
| 2 | GPT-4o May 2024 | 87 85.7 to 88.1 | Baseline: claude-3-sonnet-20240229 | May 2024 | Official leaderboard |
| 3 | Claude 3.5 Sonnet | 86 85 to 87.5 | Baseline: claude-3-sonnet-20240229 | Jun 2024 | Official leaderboard |
| 4 | | 84 81.8 to 84.6 | Baseline: claude-3-sonnet-20240229 | Aug 2024 | Official leaderboard |
| 5 | GPT-4 Turbo Apr 2024 | 81 79.7 to 82.3 | Baseline: claude-3-sonnet-20240229 | Apr 2024 | Official leaderboard |
| 6 | | 75 73.6 to 76.7 | Baseline: claude-3-sonnet-20240229 | May 2024 | Official leaderboard |
| 7 | GPT-4o mini | 73 71.7 to 74.4 | Baseline: claude-3-sonnet-20240229 | Jul 2024 | Official leaderboard |
| 8 | | 65 63.2 to 67.1 | Baseline: claude-3-sonnet-20240229 | Oct 2024 | Official leaderboard |
| 9 | | 58 56.5 to 59.9 | Baseline: claude-3-sonnet-20240229 | May 2024 | Official leaderboard |
| 10 | | 54 ±1.7 | Baseline: claude-3-sonnet-20240229 | Jul 2024 | Official leaderboard |
| 11 | Claude 3 Opus | 52 50 to 53.7 | Baseline: claude-3-sonnet-20240229 | Feb 2024 | Official leaderboard |
| 12 | Claude 3 Sonnet | 50 | Baseline: claude-3-sonnet-20240229 | Feb 2024 | Official leaderboard |
| 13 | Reka Core | 38 36 to 39.6 | Baseline: claude-3-sonnet-20240229 | Apr 2024 | Official leaderboard |
| 14 | | 33 30.9 to 34.4 | Baseline: claude-3-sonnet-20240229 | Sep 2024 | Official leaderboard |
| 15 | Claude 3 Haiku | 31 28.4 to 32.3 | Baseline: claude-3-sonnet-20240229 | Mar 2024 | Official leaderboard |
| 16 | | 20 17.7 to 20.9 | Baseline: claude-3-sonnet-20240229 | Jul 2024 | Official leaderboard |
Scores at each effort level
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.
Source: The VisionArena-Bench paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Baseline: claude-3-sonnet-20240229 | 88 | 329 | |||
| GPT-4o May 2024 | Baseline: claude-3-sonnet-20240229 | 87 | 316 | ||
| Claude 3.5 Sonnet | Baseline: claude-3-sonnet-20240229 | 86 | 262 | ||
| Baseline: claude-3-sonnet-20240229 | 84 | 367 | |||
| GPT-4 Turbo Apr 2024 | Baseline: claude-3-sonnet-20240229 | 81 | 258 | ||
| Baseline: claude-3-sonnet-20240229 | 75 | 261 | |||
| GPT-4o mini | Baseline: claude-3-sonnet-20240229 | 73 | 224 | ||
| GPT-4o May 2024 | Baseline: GPT-4-Turbo | 68 | 316 | ||
| Baseline: GPT-4-Turbo | 66 | 329 | |||
| Baseline: claude-3-sonnet-20240229 | 65 | 354 | |||
| Baseline: GPT-4-Turbo | 60 | 367 | |||
| Baseline: claude-3-sonnet-20240229 | 58 | 271 | |||
| Claude 3.5 Sonnet | Baseline: GPT-4-Turbo | 55 | 262 | ||
| Baseline: claude-3-sonnet-20240229 | 54 | 515 | |||
| Claude 3 Opus | Baseline: claude-3-sonnet-20240229 | 52 | 201 | ||
| Claude 3 Sonnet | Baseline: claude-3-sonnet-20240229 | 50 | 205 | ||
| GPT-4 Turbo Apr 2024 | Baseline: GPT-4-Turbo | 50 | 258 | ||
| Baseline: GPT-4-Turbo | 46 | 261 | |||
| GPT-4o mini | Baseline: GPT-4-Turbo | 40 | 224 | ||
| Reka Core | Baseline: claude-3-sonnet-20240229 | 38 | 252 | ||
| Baseline: claude-3-sonnet-20240229 | 33 | 457 | |||
| Claude 3 Haiku | Baseline: claude-3-sonnet-20240229 | 31 | 155 | ||
| Baseline: GPT-4-Turbo | 31 | 354 | |||
| Baseline: GPT-4-Turbo | 23 | 515 | |||
| Baseline: GPT-4-Turbo | 23 | 271 | |||
| Baseline: claude-3-sonnet-20240229 | 20 | 421 | |||
| Claude 3 Opus | Baseline: GPT-4-Turbo | 19 | 201 | ||
| Claude 3 Sonnet | Baseline: GPT-4-Turbo | 18 | 205 | ||
| Reka Core | Baseline: GPT-4-Turbo | 16 | 252 | ||
| Baseline: GPT-4-Turbo | 11 | 457 | |||
| Claude 3 Haiku | Baseline: GPT-4-Turbo | 10 | 155 | ||
| Baseline: GPT-4-Turbo | 7 | 421 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The VisionArena-Bench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official VisionArena-Bench 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 Vision
- MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy.
Qwen 3.6 Plus 86%
- MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
- CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
- MathVista In use Math reasoning in visual contexts (charts, geometry, puzzles) from 31 datasets. Answer accuracy.
Seed2.1 Pro 90.7%
- MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy.
Qwen 3.8 Max 97.7%
- MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy.
Qwen 3.8 Max 85.9%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in October 2024, for Gemini 1.5 Flash 8B. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (16 entries).
- Pass Models scored 16 models we track.
- Warning Newest score The newest came 2 years ago, in October 2024, for Gemini 1.5 Flash 8B.
- Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 1.5 Pro (#157).
- Note Headroom Scored as a rating, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 32 rows for models we track 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.