Multimodal RewardBench
Version Original
Reward models choose between two answers to a visual prompt. Accuracy measures agreement with annotated preferences across six areas.
The evaluation contains 5,000 annotated examples.
- Made by
- Meta and collaborators
- Released
- February 2025
- Kind
- General visual questions
- Level
- University
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- New versions replace it
- Run by
- Anyone
- In views
- Vision
- Strongest model tested
-
Llama 3.2 90B #179 on the capability index
- Last model added
- September 2024
- Scores refreshed
- November 20th, 2026
Claude 3.5 Sonnet (Dec 2024) tops the leaderboard with 72% Llama 3.2 90B leads with 62.4% among models released by September 2024
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. The best model we track is Llama 3.2 90B, at 62.4%. This view shows only models we track, each at its best.
The newest model with a score here came out in September 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 Claude 3.5 Sonnet (Dec 2024) 72%
- 2 Gemini 1.5 Pro (Dec 2024) 72%
- 3 GPT-4o (Dec 2024) 71.5%
- 4
Llama 3.2 90B 62.4%
- 5 Aria 57.3%
- 6
Molmo-7B-D 54.3%
- 7
Llama 3.2 11B 52.4%
- 8
LLaVA-v1.5-13B 48.9%
Source: Official Multimodal RewardBench leaderboard
- 1
Llama 3.2 90B 62.4%
- 2 Aria 57.3%
- 3
Molmo-7B-D 54.3%
- 4
Llama 3.2 11B 52.4%
- 5
LLaVA-v1.5-13B 48.9%
Source: Official Multimodal RewardBench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude 3.5 Sonnet (Dec 2024) | 72% | Official leaderboard | ||
| 2 | Gemini 1.5 Pro (Dec 2024) | 72% | Official leaderboard | ||
| 3 | GPT-4o (Dec 2024) | 71.5% | Official leaderboard | ||
| 4 | | 62.4% | Sep 2024 | Official leaderboard | |
| 5 | Aria | 57.3% | Sep 2024 | Official leaderboard | |
| 6 | | 54.3% | Sep 2024 | Official leaderboard | |
| 7 | | 52.4% | Sep 2024 | Official leaderboard | |
| 8 | | 48.9% | Oct 2023 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 62.4% | Sep 2024 | Official leaderboard | |
| 2 | Aria | 57.3% | Sep 2024 | Official leaderboard | |
| 3 | | 54.3% | Sep 2024 | Official leaderboard | |
| 4 | | 52.4% | Sep 2024 | Official leaderboard | |
| 5 | | 48.9% | Oct 2023 | Official leaderboard |
Rankings by part
We read 6 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as General Correctness, General Preference and Knowledge. Llama 3.2 90B leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.
General Correctness
5 models we track have a score here.
- 1
Llama 3.2 90B 60%
- 2 Aria 59.5%
- 3
Llama 3.2 11B ↑1 57.8%
- 4
Molmo-7B-D ↓1 56.8%
- 5
LLaVA-v1.5-13B 53.3%
General Preference
5 models we track have a score here.
- 1
Llama 3.2 90B 68.4%
- 2
Llama 3.2 11B ↑2 65.8%
- 3 Aria ↓1 63.5%
- 4
Molmo-7B-D ↓1 59.4%
- 5
LLaVA-v1.5-13B 55.2%
Knowledge
5 models we track have a score here.
- 1
Llama 3.2 90B 61.2%
- 2
Llama 3.2 11B ↑2 55.5%
- 3 Aria ↓1 55.5%
- 4
Molmo-7B-D ↓1 54.6%
- 5
LLaVA-v1.5-13B 50.5%
Reasoning
5 models we track have a score here.
- 1
Llama 3.2 90B 54.6%
- 2 Aria 52.4%
- 3
Molmo-7B-D 52.1%
- 4
LLaVA-v1.5-13B ↑1 51.3%
- 5
Llama 3.2 11B ↓1 51.2%
Safety
5 models we track have a score here.
- 1
Llama 3.2 90B 51.9%
- 2 Aria 50.2%
- 3
Molmo-7B-D 44.2%
- 4
Llama 3.2 11B 35.5%
- 5
LLaVA-v1.5-13B 34.9%
VQA
5 models we track have a score here.
- 1
Llama 3.2 90B 77.1%
- 2 Aria 64.2%
- 3
Molmo-7B-D 60.3%
- 4
Llama 3.2 11B 55.8%
- 5
LLaVA-v1.5-13B 51.8%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official Multimodal RewardBench 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 Multimodal RewardBench 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.
Versions
We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.
Source: Official leaderboards
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
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (8 entries).
- Pass Models scored 5 models we track, plus 3 other entries on its leaderboard.
- Note Newest score The newest came 2 years ago, in September 2024, for Aria. That’s expected for a test replaced by Multimodal RewardBench 2.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is Llama 3.2 90B (#179).
- Pass Headroom The best is 62.4%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 5 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.