HR-Bench
Qwen3-VL-32B-Instruct 83% leads the models we track Models answer multiple-choice questions about fine details in 4K and 8K resolution images. Results are reported separately for the 4K and 8K sets.
Labs often run it with image crop and zoom tools.
- Made by
- Wuhan University and collaborators (Wang et al.)
- Released
- August 2024
- Kind
- General visual questions
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Strongest model tested
- GPT-4o May 2024 #161 on the capability index
- Last in a launch post
- May 2026, dropped after it saturated
- Last model added
- May 2026
- Scores refreshed
- November 12th, 2026
Taichu4.0-VL tops the leaderboard with 83.5%
Best score per entry from the benchmark's own leaderboard, read on November 12th, 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 Qwen3-VL-32B-Instruct, at 83%.
MiMo-V2.5 leads with 88.5%, claims included
Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.
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:
- Claude Fable 5.2
-
Gemini 4 Carbon - GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 83.5 Taichu4.0-VL
- 83.0 Qwen3-VL-32B-Instruct
- 80.9 Qwen3-VL-30B-A3B-Instruct
- 68.6 MiMo-VL-SFT 7B
- 66.9
InternVL2-Llama3-76B - 66.6 Qwen2-VL-7B-Instruct
- 66.0 GPT-4o
- 64.8 Gemini 1.5 Flash (snapshot unspecified)
- 62.3 InternVL-1.5-26B
- 61.3 Qwen2-VL-2B-Instruct
- 55.5 Qwen-VL-Max
- 54.6 Xcomposer2-4kHD-7B
- 50.3 LLaVA-HR-X-13B
- 50.2 LLaVA-1.6-34B
- 49.8 Qwen-VL-Plus
- 46.8 LLaVA-HR-X-7B
- 1 Taichu4.0-VL 83.5%
- 2 Qwen3-VL-32B-Instruct 83%
- 3 Qwen3-VL-30B-A3B-Instruct 80.9%
- 4 MiMo-VL-SFT 7B 68.6%
- 5 InternVL2-Llama3-76B 66.9%
- 6 Qwen2-VL-7B-Instruct 66.6%
- 7 GPT-4o May 2024 66%
- 8 Gemini 1.5 Flash (snapshot unspecified) 64.8%
- 9 InternVL-1.5-26B 62.3%
- 10 Qwen2-VL-2B-Instruct 61.3%
- 11 Qwen-VL-Max 55.5%
- 12 Xcomposer2-4kHD-7B 54.6%
- 13 LLaVA-HR-X-13B 50.3%
- 14 LLaVA-1.6-34B 50.2%
- 15 Qwen-VL-Plus 49.8%
- 16 LLaVA-HR-X-7B 46.8%
Source: Official HR-Bench leaderboard
- 1 MiMo-V2.5 88.5%
- 2
Step 3.7 Flash 87.7% - 3 Qwen3-VL-32B-Instruct 83%
- 4 Qwen3-VL-30B-A3B-Instruct 80.9%
- 5 InternVL2-Llama3-76B 66.9%
- 6 Qwen2-VL-7B-Instruct 66.6%
- 7 GPT-4o May 2024 66%
- 8 Qwen2-VL-2B-Instruct 61.3%
- 9 Qwen-VL-Max 55.5%
- 10 Qwen-VL-Plus 49.8%
Sources: Labs’ model cards and official HR-Bench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Taichu4.0-VL | 83.5% | Official leaderboard | ||
| 2 | Qwen3-VL-32B-Instruct | 83% | Oct 2025 | Official leaderboard | |
| 3 | Qwen3-VL-30B-A3B-Instruct | 80.9% | Oct 2025 | Official leaderboard | |
| 4 | MiMo-VL-SFT 7B | 68.6% | Official leaderboard | ||
| 5 | InternVL2-Llama3-76B | 66.9% | DC² | Jul 2024 | Official leaderboard |
| 6 | Qwen2-VL-7B-Instruct | 66.6% | Aug 2024 | Official leaderboard | |
| 7 | GPT-4o May 2024 | 66% | May 2024 | Official leaderboard | |
| 8 | Gemini 1.5 Flash (snapshot unspecified) | 64.8% | Official leaderboard | ||
| 9 | InternVL-1.5-26B | 62.3% | DC² | Official leaderboard | |
| 10 | Qwen2-VL-2B-Instruct | 61.3% | Aug 2024 | Official leaderboard | |
| 11 | Qwen-VL-Max | 55.5% | Jan 2024 | Official leaderboard | |
| 12 | Xcomposer2-4kHD-7B | 54.6% | Official leaderboard | ||
| 13 | LLaVA-HR-X-13B | 50.3% | Official leaderboard | ||
| 14 | LLaVA-1.6-34B | 50.2% | Official leaderboard | ||
| 15 | Qwen-VL-Plus | 49.8% | Jan 2024 | Official leaderboard | |
| 16 | LLaVA-HR-X-7B | 46.8% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | MiMo-V2.5 | 88.5% | Apr 2026 | Lab model card | |
| 2 | Step 3.7 Flash | 87.7% | May 2026 | Lab model card | |
| 3 | Qwen3-VL-32B-Instruct | 83% | Oct 2025 | Official leaderboard | |
| 4 | Qwen3-VL-30B-A3B-Instruct | 80.9% | Oct 2025 | Official leaderboard | |
| 5 | InternVL2-Llama3-76B | 66.9% | DC² | Jul 2024 | Official leaderboard |
| 6 | Qwen2-VL-7B-Instruct | 66.6% | Aug 2024 | Official leaderboard | |
| 7 | GPT-4o May 2024 | 66% | May 2024 | Official leaderboard | |
| 8 | Qwen2-VL-2B-Instruct | 61.3% | Aug 2024 | Official leaderboard | |
| 9 | Qwen-VL-Max | 55.5% | Jan 2024 | Official leaderboard | |
| 10 | Qwen-VL-Plus | 49.8% | Jan 2024 | Official leaderboard |
Labs report scores for 2 models in their cards. Switch the scores above to rank them together.
3 models have held number one on the official leaderboard since its release in August 2024 4 models have held number one with claims included since its release in August 2024
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official HR-Bench leaderboard
Xiaomi Corp’s own number for MiMo-V2.5 passed 85% in April 2026
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Sources: Official HR-Bench leaderboard and labs’ model cards
Labs headline it less often now, last in May 2026
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| May 29, 2026 | Step 3.7 Flash | 87.7 | Mean of HR-Bench 4K 89.13 / HR-Bench 8K 86.34 (with Python) |
| Apr 22, 2026 | MiMo-V2.5 | 88.5 | HR-Bench (4k) |
One system card reports it
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 26, 2025 | Qwen3-VL-235B-A22B technical report | HRBench4K 84.3, HRBench8K 76.6, no overall (Table 2) | Capabilities |
More in Vision
- MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. GPT-5.1 85.4%
- 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%
- 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%
- ChartQA In use Models answer human-written and generated questions about chart images that need visual and arithmetic reasoning. Scored by relaxed accuracy with a 5 percent numeric tolerance.
Command A Vision 90.9%