RBench-M
Models answer 665 graduate-level multimodal reasoning questions in English and Chinese. Top-1 accuracy compares selected answers with the answer key.
This is the multimodal track of the multidisciplinary RBench. It is distinct from the existing rbench record about video generation. English and Chinese results are separate language tracks.
- Made by
- Meng-Hao Guo and collaborators
- Released
- May 2025
- Kind
- General visual questions
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Subject
- Mathematics, Physical sciences, Chemical sciences, Biological sciences, Electrical, electronic and information engineering
- Interaction
- Single turn
- Items from
- Another benchmark
- Refresh
- One fixed set
- Run by
- Anyone
- Based on
- A version of another benchmark
- In views
- Vision, Biology and medicine, Chemistry and materials
- Strongest model tested
- GPT-5 #71 on the capability index
- Last model added
- October 2025
- Scores refreshed
- November 20th, 2026
In the authors’ words
Reasoning stands as a cornerstone of intelligence, enabling the synthesis of existing knowledge to solve complex problems. Despite remarkable progress, existing reasoning benchmarks often fail to rigorously evaluate the nuanced reasoning capabilities required for complex, real-world problemsolving, particularly in multi-disciplinary and multimodal contexts. In this paper, we introduce a graduate-level, multi-disciplinary, EnglishChinese benchmark, dubbed as Reasoning Bench (R-Bench), for assessing the reasoning capability of both language and multimodal models. RBench spans 1,094 questions across 108 subjects for language model evaluation and 665 questions across 83 subjects for multimodal model testing in both English and Chinese. These questions are meticulously curated to ensure rigorous difficulty calibration, subject balance, and crosslinguistic alignment, enabling the assessment to be an Olympiad-level multi-disciplinary benchmark. We evaluate widely used models, including OpenAI o1, GPT-4o, DeepSeek-R1, etc. Experimental results indicate that advanced models perform poorly on complex reasoning, especially multimodal reasoning. Even the top-performing model OpenAI o1 achieves only 53.2% accuracy on our multimodal evaluation. Data and code are made publicly available at here.
GPT-5 leads with 69.9% among models released by October 2025 GPT-5 leads with 69.9% among models released by October 2025
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. 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 October 2025, 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
- 69.9 GPT-5
- 63.3
Doubao-Seed-1.6
- 53.1 o1
- 41.0 GLM-4.5V
- 40.2
Doubao-1.5-pro
- 39.0 Claude 3.5 Sonnet (October 2024)
- 37.8 JT-VL-Chat-Thinking-20251015
- 36.8
Qwen3-VL-30B-A3B-Instruct
- 35.7
Gemini 1.5 Pro
- 33.3 GPT-4o
- 25.4 Qwen2-VL-72B
- 23.7
LLaVA-OneVision 7B
- 23.1
DeepSeek-VL2
- 21.0 Qwen2-VL-7B-Instruct
- 19.3
Llama 3.2 11B
- 1 GPT-5 69.9%
- 2
Doubao-Seed-1.6 63.3%
- 3 o1 53.1%
- 4 GLM-4.5V 41%
- 5
Doubao-1.5-pro 40.2%
- 6 Claude 3.5 Sonnet (October 2024) 39%
- 7 JT-VL-Chat-Thinking-20251015 37.8%
- 8
Qwen3-VL-30B-A3B-Instruct 36.8%
- 9
Gemini 1.5 Pro Feb 2024 35.7%
- 10 GPT-4o Nov 2024 33.3%
- 11 Qwen2-VL-72B 25.4%
- 12
LLaVA-OneVision 7B 23.7%
- 13
DeepSeek-VL2 23.1%
- 14 Qwen2-VL-7B-Instruct 21%
- 15
Llama 3.2 11B 19.3%
Source: Official RBench-M leaderboard
- 1 GPT-5 69.9%
- 2
Doubao-Seed-1.6 63.3%
- 3 o1 53.1%
- 4 GLM-4.5V 41%
- 5
Doubao-1.5-pro 40.2%
- 6 Claude 3.5 Sonnet (October 2024) 39%
- 7
Qwen3-VL-30B-A3B-Instruct 36.8%
- 8
Gemini 1.5 Pro Feb 2024 35.7%
- 9 GPT-4o Nov 2024 33.3%
- 10
LLaVA-OneVision 7B 23.7%
- 11
DeepSeek-VL2 23.1%
- 12 Qwen2-VL-7B-Instruct 21%
- 13
Llama 3.2 11B 19.3%
Source: Official RBench-M leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 69.9% | Aug 2025 | Official leaderboard | |
| 2 | | 63.3% | Jun 2025 | Official leaderboard | |
| 3 | o1 | 53.1% | Dec 2024 | Official leaderboard | |
| 4 | GLM-4.5V | 41% | Aug 2025 | Official leaderboard | |
| 5 | | 40.2% | Jan 2025 | Official leaderboard | |
| 6 | Claude 3.5 Sonnet (October 2024) | 39% | Oct 2024 | Official leaderboard | |
| 7 | JT-VL-Chat-Thinking-20251015 | 37.8% | Official leaderboard | ||
| 8 | | 36.8% | Oct 2025 | Official leaderboard | |
| 9 | | 35.7% | Feb 2024 | Official leaderboard | |
| 10 | GPT-4o Nov 2024 | 33.3% | Nov 2024 | Official leaderboard | |
| 11 | Qwen2-VL-72B | 25.4% | Official leaderboard | ||
| 12 | | 23.7% | Aug 2024 | Official leaderboard | |
| 13 | | 23.1% | Dec 2024 | Official leaderboard | |
| 14 | Qwen2-VL-7B-Instruct | 21% | Aug 2024 | Official leaderboard | |
| 15 | | 19.3% | Sep 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 69.9% | Aug 2025 | Official leaderboard | |
| 2 | | 63.3% | Jun 2025 | Official leaderboard | |
| 3 | o1 | 53.1% | Dec 2024 | Official leaderboard | |
| 4 | GLM-4.5V | 41% | Aug 2025 | Official leaderboard | |
| 5 | | 40.2% | Jan 2025 | Official leaderboard | |
| 6 | Claude 3.5 Sonnet (October 2024) | 39% | Oct 2024 | Official leaderboard | |
| 7 | | 36.8% | Oct 2025 | Official leaderboard | |
| 8 | | 35.7% | Feb 2024 | Official leaderboard | |
| 9 | GPT-4o Nov 2024 | 33.3% | Nov 2024 | Official leaderboard | |
| 10 | | 23.7% | Aug 2024 | Official leaderboard | |
| 11 | | 23.1% | Dec 2024 | Official leaderboard | |
| 12 | Qwen2-VL-7B-Instruct | 21% | Aug 2024 | Official leaderboard | |
| 13 | | 19.3% | Sep 2024 | Official leaderboard |
Rankings by part
We read 2 parts from the benchmark’s own leaderboard: RBench-M (English) and RBench-M (Chinese). Each ranks models on one part of the test. GPT-5 leads both. Arrows show how far a model moved from its overall rank among the models we track.
RBench-M (English)
13 models we track have a score here.
- 1 GPT-5 68.4%
- 2
Doubao-Seed-1.6 62.4%
- 3 o1 53.2%
- 4 GLM-4.5V 40.8%
- 5 Claude 3.5 Sonnet (October 2024) ↑1 39.7%
- 6
Doubao-1.5-pro ↓1 37.9%
- 7
Qwen3-VL-30B-A3B-Instruct 36.5%
- 8
Gemini 1.5 Pro Feb 2024 35.9%
- 9 GPT-4o Nov 2024 33.4%
- 10
LLaVA-OneVision 7B 23.8%
RBench-M (Chinese)
13 models we track have a score here.
- 1 GPT-5 71.3%
- 2
Doubao-Seed-1.6 64.2%
- 3 o1 53%
- 4
Doubao-1.5-pro ↑1 42.4%
- 5 GLM-4.5V ↓1 41.2%
- 6 Claude 3.5 Sonnet (October 2024) 38.3%
- 7
Qwen3-VL-30B-A3B-Instruct 37%
- 8
Gemini 1.5 Pro Feb 2024 35.5%
- 9 GPT-4o Nov 2024 33.2%
- 10
DeepSeek-VL2 ↑1 24.4%
3 models have held number one on the official leaderboard since its release in May 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official RBench-M leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official RBench-M 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
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 13 months ago, in October 2025, for Qwen3-VL-30B-A3B-Instruct. 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 (15 entries).
- Pass Models scored 13 models we track, plus 2 other entries on its leaderboard.
- Warning Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-30B-A3B-Instruct.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5 (#71).
- Pass Headroom The best is 69.9%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on evalmodels.github.io: 13 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.