MM-AlignBench
Models answer open-ended questions about 252 human-annotated images. GPT-4o compares their answers with Claude-Sonnet-3 reference answers to assess alignment with human preferences.
The evaluation set is separate from the OmniAlign-V training datasets. The star count belongs to the combined project.
- Released
- February 2025
- 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
- Written by crowdworkers
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision
- Last model added
- November 2024
- Scores refreshed
- November 19th, 2026
In the authors’ words
Recent advancements in open-source multi-modal large language models (MLLMs) have primarily focused on enhancing foundational capabilities, leaving a significant gap in human preference alignment. This paper introduces OmniAlign-V, a comprehensive dataset of 200K high-quality training samples featuring diverse images, complex questions, and varied response formats to improve MLLMs' alignment with human preferences. We also present MM-AlignBench, a human-annotated benchmark specifically designed to evaluate MLLMs' alignment with human values. Experimental results show that finetuning MLLMs with OmniAlign-V, using Supervised Fine-Tuning (SFT) or Direct Preference Optimization (DPO), significantly enhances human preference alignment while maintaining or enhancing performance on standard VQA benchmarks, preserving their fundamental capabilities. Our datasets, benchmark, code and checkpoints have been released at https://github.com/PhoenixZ810/OmniAlign-V.
Claude3.5V-Sonnet tops the leaderboard with 84.9% GPT-4V leads with 82.5% among models released by November 2024
Best score per entry from the benchmark's own leaderboard, read on November 19th, 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. The best model we track is GPT-4V, at 82.5%. This view shows only models we track, each at its best.
The newest model with a score here came out in November 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
- 84.9 Claude3.5V-Sonnet
- 82.5 GPT-4V
- 81.3 GPT-4o (snapshot unspecified)
- 77.0 GeminiFlash1.5-002
- 74.2 LLaVANext-OA-32B-DPO
- 62.3 LLaVANext-OA-32B
- 61.5 Qwen2VL-72B
- 50.0 Claude-3V-Sonnet
- 44.4 InternVL2-72B
- 44.4 Qwen2-VL-7B-Instruct
- 40.1
InternVL2-8B-MPO
- 31.3
InternVL2-8B
- 27.8 LLaMA3.2-Vision-11B
- 26.6 LLaVANext-Qwen32B
- 23.8
LLaVA-OneVision 7B
- 12.7
MiniCPM-Llama3-V 2.5
- 7.50 Xcomposer2.5-7B
- 2.70
Idefics3-8B-Llama3
- 1 Claude3.5V-Sonnet 84.9%
- 2 GPT-4V 82.5%
- 3 GPT-4o (snapshot unspecified) 81.3%
- 4 GeminiFlash1.5-002 77%
- 5 LLaVANext-OA-32B-DPO 74.2%
- 6 LLaVANext-OA-32B 62.3%
- 7 Qwen2VL-72B 61.5%
- 8 Claude-3V-Sonnet 50%
- 9 InternVL2-72B 44.4%
- 10 Qwen2-VL-7B-Instruct 44.4%
- 11
InternVL2-8B-MPO 40.1%
- 12
InternVL2-8B 31.3%
- 13 LLaMA3.2-Vision-11B 27.8%
- 14 LLaVANext-Qwen32B 26.6%
- 15
LLaVA-OneVision 7B 23.8%
- 16
MiniCPM-Llama3-V 2.5 12.7%
- 17 Xcomposer2.5-7B 7.50%
- 18
Idefics3-8B-Llama3 2.70%
Source: Official MM-AlignBench leaderboard
- 1 GPT-4V 82.5%
- 2 Qwen2-VL-7B-Instruct 44.4%
- 3
InternVL2-8B-MPO 40.1%
- 4
InternVL2-8B 31.3%
- 5
LLaVA-OneVision 7B 23.8%
- 6
MiniCPM-Llama3-V 2.5 12.7%
- 7
Idefics3-8B-Llama3 2.70%
Source: Official MM-AlignBench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude3.5V-Sonnet | 84.9% | Official leaderboard | ||
| 2 | GPT-4V | 82.5% | Nov 2023 | Official leaderboard | |
| 3 | GPT-4o (snapshot unspecified) | 81.3% | Official leaderboard | ||
| 4 | GeminiFlash1.5-002 | 77% | Official leaderboard | ||
| 5 | LLaVANext-OA-32B-DPO | 74.2% | Official leaderboard | ||
| 6 | LLaVANext-OA-32B | 62.3% | Official leaderboard | ||
| 7 | Qwen2VL-72B | 61.5% | Official leaderboard | ||
| 8 | Claude-3V-Sonnet | 50% | Official leaderboard | ||
| 9 | InternVL2-72B | 44.4% | Official leaderboard | ||
| 10 | Qwen2-VL-7B-Instruct | 44.4% | Aug 2024 | Official leaderboard | |
| 11 | | 40.1% | Nov 2024 | Official leaderboard | |
| 12 | | 31.3% | Jul 2024 | Official leaderboard | |
| 13 | LLaMA3.2-Vision-11B | 27.8% | Official leaderboard | ||
| 14 | LLaVANext-Qwen32B | 26.6% | Official leaderboard | ||
| 15 | | 23.8% | Aug 2024 | Official leaderboard | |
| 16 | | 12.7% | May 2024 | Official leaderboard | |
| 17 | Xcomposer2.5-7B | 7.50% | Official leaderboard | ||
| 18 | | 2.70% | Aug 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4V | 82.5% | Nov 2023 | Official leaderboard | |
| 2 | Qwen2-VL-7B-Instruct | 44.4% | Aug 2024 | Official leaderboard | |
| 3 | | 40.1% | Nov 2024 | Official leaderboard | |
| 4 | | 31.3% | Jul 2024 | Official leaderboard | |
| 5 | | 23.8% | Aug 2024 | Official leaderboard | |
| 6 | | 12.7% | May 2024 | Official leaderboard | |
| 7 | | 2.70% | Aug 2024 | Official leaderboard |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official MM-AlignBench 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 MM-AlignBench 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 2 years ago, in November 2024, for InternVL2-8B-MPO. 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 (18 entries).
- Pass Models scored 7 models we track, plus 11 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in November 2024, for InternVL2-8B-MPO.
- Note Frontier models None of its models are on the capabilities index.
- Pass Headroom The best is 82.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on huggingface.co: 7 rows for models we track and 11 other entries at the last read.
- Pass Last read November 19th, 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.