VSI-Bench
InternVL3.5-241B-A28B 69.5% leads since August 2025 InternVL3.5-241B-A28B 69.5% leads since August 2025 Models watch egocentric videos of 288 indoor scenes and answer over 5,000 questions on object counts, distances, sizes, room layout and route planning.
Mixes multiple-choice questions scored by accuracy with numeric answers scored by mean relative accuracy. Videos come from ScanNet, ScanNet++ and ARKitScenes.
- Made by
- New York University (VisionX, Saining Xie) with Stanford and Yale
- Released
- December 2024
- Kind
- Spatial reasoning and counting
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Last in a launch post
- January 2026, dropped before it saturated
- Last model added
- October 2026
- Scores refreshed
- November 12th, 2026
In the authors’ words
Humans possess the visual-spatial intelligence to remember spaces from sequential visual observations. However, can Multimodal Large Language Models (MLLMs) trained on million-scale video datasets also “think in space” from videos? We present a novel video-based visual-spatial intelligence benchmark (VSI-Bench) of over 5,000 question-answer pairs, and find that MLLMs exhibit competitive - though subhuman - visual-spatial intelligence. We probe models to express how they think in space both linguistically and visually and find that while spatial reasoning capabilities remain the primary bottleneck for MLLMs to reach higher benchmark performance, local world models and spatial awareness do emerge within these models. Notably, prevailing linguistic reasoning techniques (e.g., chain-of-thought, self-consistency, tree-of-thoughts) fail to improve performance, whereas explicitly generating cognitive maps during question-answering enhances MLLMs' spatial distance ability.
InternVL3.5-241B-A28B leads with 69.5%
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.
LLaVA-OneVision 72B leads with 40.2% across all independent runs
Best score per model across the official leaderboard, Epoch AI’s own runs and any board Epoch AI copied. Each bar and row names its source.
InternVL3.5-241B-A28B leads with 69.5%, 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.
- 1 InternVL3.5-241B-A28B 69.5%
- 2
SenseNova-SI-1.3-InternVL3-8B 68.6% - 3
Qwen3-VL-235B-A22B 60% - 4
Cosmos3-Edge 59.2% - 5 Cosmos3-Nano 54.9%
- 6 Nemotron 3.5 Super VL 48.1%
- 7
gemini-robotics-er-1.5-preview 45.8%
Source: Labs’ model cards
- 1 LLaVA-OneVision 72B 40.2%
- 2 InternVL2-8B 37.5%
- 3 InternVL2-40B 37%
- 4
LLaVA-OneVision 7B 32.4% - 5 VILA1.5-40B 31.2%
- 6 LongVA-7B 29.2%
- 7 Llama-3-VILA1.5-8B 28.9%
- 8 LLaVA-OneVision 0.5B 28%
- 9 InternVL2-2B 26.5%
Source: Official VSI-Bench leaderboard
- 69.5 InternVL3.5-241B-A28B
- 68.6 SenseNova-SI-1.3-InternVL3-8B
- 60.0 Qwen3-VL-235B-A22B
- 59.2 Cosmos3-Edge
- 54.9 Cosmos3-Nano
- 48.1 Nemotron 3.5 Super VL
- 45.8 gemini-robotics-er-1.5-preview
- 40.2 LLaVA-OneVision 72B
- 37.5 InternVL2-8B
- 37.0 InternVL2-40B
- 32.4 LLaVA-OneVision 7B
- 31.2 VILA1.5-40B
- 29.2 LongVA-7B
- 28.9 Llama-3-VILA1.5-8B
- 28.0 LLaVA-OneVision 0.5B
- 26.5 InternVL2-2B
- 1 InternVL3.5-241B-A28B 69.5%
- 2 SenseNova-SI-1.3-InternVL3-8B 68.6%
- 3 Qwen3-VL-235B-A22B 60%
- 4 Cosmos3-Edge 59.2%
- 5 Cosmos3-Nano 54.9%
- 6 Nemotron 3.5 Super VL 48.1%
- 7 gemini-robotics-er-1.5-preview 45.8%
- 8 LLaVA-OneVision 72B 40.2%
- 9 InternVL2-8B 37.5%
- 10 InternVL2-40B 37%
- 11 LLaVA-OneVision 7B 32.4%
- 12 VILA1.5-40B 31.2%
- 13 LongVA-7B 29.2%
- 14 Llama-3-VILA1.5-8B 28.9%
- 15 LLaVA-OneVision 0.5B 28%
- 16 InternVL2-2B 26.5%
Sources: Labs’ model cards and official VSI-Bench leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | InternVL3.5-241B-A28B | 69.5% | Aug 2025 | Lab model card | |
| 2 | SenseNova-SI-1.3-InternVL3-8B | 68.6% | Jan 2026 | Lab model card | |
| 3 | Qwen3-VL-235B-A22B | 60% | Nov 2025 | Lab model card | |
| 4 | Cosmos3-Edge | 59.2% | May 2026 | Lab model card | |
| 5 | Cosmos3-Nano | 54.9% | May 2026 | Lab model card | |
| 6 | Nemotron 3.5 Super VL | 48.1% | Oct 2026 | Lab model card | |
| 7 | gemini-robotics-er-1.5-preview | 45.8% | Sep 2025 | Lab model card |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | LLaVA-OneVision 72B | 40.2% | Aug 2024 | Official leaderboard | |
| 2 | InternVL2-8B | 37.5% | Jul 2024 | Official leaderboard | |
| 3 | InternVL2-40B | 37% | Jul 2024 | Official leaderboard | |
| 4 | LLaVA-OneVision 7B | 32.4% | Aug 2024 | Official leaderboard | |
| 5 | VILA1.5-40B | 31.2% | Apr 2024 | Official leaderboard | |
| 6 | LongVA-7B | 29.2% | Jun 2024 | Official leaderboard | |
| 7 | Llama-3-VILA1.5-8B | 28.9% | Apr 2024 | Official leaderboard | |
| 8 | LLaVA-OneVision 0.5B | 28% | Aug 2024 | Official leaderboard | |
| 9 | InternVL2-2B | 26.5% | Jul 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | InternVL3.5-241B-A28B | 69.5% | Aug 2025 | Lab model card | |
| 2 | SenseNova-SI-1.3-InternVL3-8B | 68.6% | Jan 2026 | Lab model card | |
| 3 | Qwen3-VL-235B-A22B | 60% | Nov 2025 | Lab model card | |
| 4 | Cosmos3-Edge | 59.2% | May 2026 | Lab model card | |
| 5 | Cosmos3-Nano | 54.9% | May 2026 | Lab model card | |
| 6 | Nemotron 3.5 Super VL | 48.1% | Oct 2026 | Lab model card | |
| 7 | gemini-robotics-er-1.5-preview | 45.8% | Sep 2025 | Lab model card | |
| 8 | LLaVA-OneVision 72B | 40.2% | Aug 2024 | Official leaderboard | |
| 9 | InternVL2-8B | 37.5% | Jul 2024 | Official leaderboard | |
| 10 | InternVL2-40B | 37% | Jul 2024 | Official leaderboard | |
| 11 | LLaVA-OneVision 7B | 32.4% | Aug 2024 | Official leaderboard | |
| 12 | VILA1.5-40B | 31.2% | Apr 2024 | Official leaderboard | |
| 13 | LongVA-7B | 29.2% | Jun 2024 | Official leaderboard | |
| 14 | Llama-3-VILA1.5-8B | 28.9% | Apr 2024 | Official leaderboard | |
| 15 | LLaVA-OneVision 0.5B | 28% | Aug 2024 | Official leaderboard | |
| 16 | InternVL2-2B | 26.5% | Jul 2024 | Official leaderboard |
The benchmark's own leaderboard has 15 entries, 9 of them models we track. Switch the scores above to rank them together.
Number one over time Number one over time 2 models have held number one with claims included since its release in December 2024
Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.
Sources: Official VSI-Bench leaderboard and labs’ model cards
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Sources: Official VSI-Bench leaderboard and labs’ model cards
Labs headline it less often now, last in January 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 |
|---|---|---|---|
| Jan 8, 2026 | SenseNova-SI-1.3-InternVL3-8B | 68.6 | VSI |
| Aug 25, 2025 | InternVL3.5-241B-A28B | 69.5 | VSI-Bench |
6 system cards report it, from 4 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 |
|---|---|---|---|---|
| Oct 15, 2026 | Nemotron 3.5 Super VL technical report | 48.1 | VSI-Bench-U, SpatialClaw harness, with agentic vision training, Table 7 | Capabilities |
| May 31, 2026 | Cosmos3-Nano technical report | 54.9 | VSI-Bench, Cosmos3-Nano 8B column, Table 10 | Capabilities |
| May 31, 2026 | Cosmos3-Edge technical report | 59.2 | VSI-Bench, Cosmos3-Edge 2B column, Table 10 | Capabilities |
| Nov 26, 2025 | Qwen3-VL-235B-A22B technical report | 60 | VSI-Bench (Table 2) | Capabilities |
| Sep 25, 2025 | gemini-robotics-er-1.5-preview technical report | 45.8 | VSI-Bench (Table 19, thinking on). Thinking off 39.9 | Capabilities |
| Aug 25, 2025 | InternVL3.5-241B-A28B technical report | 69.5 | VSI-Bench (Table 2, Table 11) | 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%