MuirBench
Seed2.0 Pro 81.8% leads since Feb 2026 Fading Models answer multiple choice questions over sets of several images, including unanswerable variants. Scored by accuracy.
In the authors’ words
We introduce MuirBench, a comprehensive benchmark that focuses on robust multi-image understanding capabilities of multimodal LLMs. MuirBench consists of 12 diverse multi-image tasks (e.g., scene understanding, ordering) that involve 10 categories of multi-image relations (e.g., multiview, temporal relations). Comprising 11,264 images and 2,600 multiple-choice questions, MuirBench is created in a pairwise manner, where each standard instance is paired with an unanswerable variant that has minimal semantic differences, in order for a reliable assessment. Evaluated upon 20 recent multi-modal LLMs, our results reveal that even the best-performing models like GPT-4o and Gemini Pro find it challenging to solve MuirBench, achieving 68.0% and 49.3% in accuracy. Open-source multimodal LLMs trained on single images can hardly generalize to multi-image questions, hovering below 33.3% in accuracy. These results highlight the importance of MuirBench in encouraging the community to develop multimodal LLMs that can look beyond a single image, suggesting potential pathways for future improvements.
- Level
- Everyday Most adults, without training
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Multiple choice
Seed2.0 Pro leads with 81.8% among models released by Aug 2026
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in Aug 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 16 more
- 1 Seed2.0 Pro 81.8%
- 2
Qwen3-VL-235B-A22B 80.1% - 3 Seed1.8 78.7%
- 4
LongCat-Flash-Omni 77.1% - 5 GLM-4.6V 77.1%
- 6 GLM-4.5V 75.3%
- 7 GLM-4.1V-9B-Thinking 74.7%
- 8 Qwen2.5-VL-72B 70.7%
- 9 Qwen2.5-Omni-7B 59.2%
- 10 LFM2.5-VL-3B 58.3%
- 11
InternVL3.5-241B-A28B 47.5% - 12
Nemotron Nano 12B v2 VL 44.2%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Seed2.0 Pro | 81.8% | Feb 2026 | Model card | |
| 2 | Qwen3-VL-235B-A22B | 80.1% | Sep 2025 | Model card | |
| 3 | Seed1.8 | 78.7% | Dec 2025 | Model card | |
| 4 | LongCat-Flash-Omni | 77.1% | Oct 2025 | Model card | |
| 5 | GLM-4.6V | 77.1% | Dec 2025 | Model card | |
| 6 | GLM-4.5V | 75.3% | Aug 2025 | Model card | |
| 7 | GLM-4.1V-9B-Thinking | 74.7% | Jul 2025 | Model card | |
| 8 | Qwen2.5-VL-72B | 70.7% | Feb 2025 | Model card | |
| 9 | Qwen2.5-Omni-7B | 59.2% | Mar 2025 | Model card | |
| 10 | LFM2.5-VL-3B | 58.3% | Aug 2026 | Model card | |
| 11 | InternVL3.5-241B-A28B | 47.5% | Aug 2025 | Model card | |
| 12 | Nemotron Nano 12B v2 VL | 44.2% | Nov 2025 | Model card |
5 models have held number one since its release in Jun 2024
Lab claims count from the day the card came out, independent runs from the model’s release.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs headline it less often now, last in Aug 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 |
|---|---|---|---|
| 12 Aug 2026 | LFM2.5-VL-3B | 58.3 | MuirBench |
| 18 Dec 2025 | Seed1.8 | 78.7 | MUIRBench |
| 8 Dec 2025 | GLM-4.6V | 77.1 | MUIRBENCH |
| 31 Oct 2025 | LongCat-Flash-Omni | 77.1 | MuirBench |
| 23 Sep 2025 | Qwen3-VL-235B-A22B | 80.1 | MUIRBENCH |
| 11 Aug 2025 | GLM-4.5V | 75.3 | MUIRBENCH |
| 1 Jul 2025 | GLM-4.1V-9B-Thinking | 74.7 | MUIRBENCH |
| 27 Mar 2025 | Qwen2.5-Omni-7B | 59.2 | MuirBench |
11 system cards report it, from 6 labs
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 |
|---|---|---|---|---|
| 21 Mar 2026 | Seed1.8 model card | 78.7 | MUIRBench (Table 2) | Capabilities |
| 14 Feb 2026 | Seed2.0 Pro model card | 81.8 | MUIRBench (Table 8) | Capabilities |
| 1 Jan 2026 | GLM-4.6V technical report | 77.1 | MUIRBENCH, GLM-4.6V thinking (Table 2) | Capabilities |
| 26 Nov 2025 | Qwen3-VL-235B-A22B technical report | 80.1 | MUIRBENCH (Table 2) | Capabilities |
| 6 Nov 2025 | Nemotron Nano 12B v2 VL technical report | 44.2 | MUIRBench, reasoning on (Table 3). Reasoning off 33.2 | Capabilities |
| 31 Oct 2025 | LongCat-Flash-Omni technical report | 77.1 | MuirBench (Table 5) | Capabilities |
| 25 Aug 2025 | InternVL3.5-241B-A28B technical report | 47.5 | MuirBench (Table 5), checked on the rendered page | Capabilities |
| 13 Aug 2025 | GLM-4.5V technical report | 75.3 | MUIRBENCH, GLM-4.5V thinking (Table 2) | Capabilities |
| 1 Jul 2025 | GLM-4.1V-9B-Thinking technical report | 74.7 | MUIRBENCH, GLM-4.1V-9B-Thinking thinking (Table 2) | Capabilities |
| 26 Mar 2025 | Qwen2.5-Omni-7B technical report | 59.2 | MuirBench (Table 5) | Capabilities |
| 19 Feb 2025 | Qwen2.5-VL-72B technical report | 70.7 | MuirBench (Table 3) | 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. Kimi K3.1 Pro 84.1%
- 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%
- 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% - DocVQA In use Models answer 50,000 questions about 12,000 scanned document images by reading their text and layout. Scoring is ANLS, a string similarity to reference answers. Seed1.5-VL 96.9%