README.md

---
license: cc
task_categories:
- multiple-choice
- visual-question-answering
- video-text-to-text
language:
- en
size_categories:
- 1K<n<10K
---

# MMSI-Video-Bench: A Holistic Benchmark for Video-Based Spatial Intelligence
[**🌐 Homepage**](https://rbler1234.github.io/MMSI-VIdeo-Bench.github.io/)  | [**📑 Paper**](https://arxiv.org/abs/2512.10863) | [**📖 Code**](https://github.com/InternRobotics/MMSI-Video-Bench)
</div>


<!-- contents with emoji -->

## 🔔 News
🔥[2025-12]: Our MMSI-Video-Bench has been integrated into [VLMEvalKit](https://github.com/open-compass/VLMEvalKit).
🔥[2025-12]: We released our paper, benchmark, and evaluation codes.


## 📊 Data Details

All of our data is available on [Hugging Face](https://huggingface.co/datasets/rbler/MMSI-Video-Bench) and includes the following components:

🎥 **Video Data** (`videos.zip`): Contains the video clip file (.mp4) corresponding to each sample. This file is generally not required for most models.

🎥 **Frame Data** (`frames.zip`): Contains the frames (.jpg) extracted from each sample's video at the **base sampling rate**. This rate ensures no key information loss during sampling.  Each frame file is named using the format `{timestamp}_frame_{base_interval}_{image_id}` (e.g., 00:06.00_frame_1.50_4), where the timestamp, also shown on the **top-left corner** of the frame, indicates its **capture time in the original recording**.

🖼️ **Reference Image Data** (`ref_images.zip`): Contains the auxiliary images referenced in the questions for each sample.

📝 **Text Annotation** (`mmsivideo.json`)：This file contains the annotation information for MMSI-Video-Bench. All time references in the questions correspond to the capture time in the original recording and **align with** the timestamp flag on each frame. Key fields include:

```
{
  "ref_images": [Paths to auxiliary images referenced in the question,...],
  "video_list": [
    {
      "path": Video clip file path,
      "start": Timestamp (in seconds) of the first frame of the video clip in the original recording,
      "end": Timestamp (in seconds) of the last frame of the video clip in the original recording,
      "base_fps": Base sampling rate
    },
    ...
  ],
  "frames_list": [[Paths to frames sampled at the base sampling rate,...],...],
  "system_prompt": "...",
  "task_prompt": Task-specific prompt,
  "user_prompt": Question text, with <video> as a placeholder for video and <image> for auxiliary images,
  "format_prompt": Output format requirements,
  "ground_truth": Correct answer
}
```


Unless otherwise specified, the model input generally consists of:
`system_prompt + task_prompt + user_prompt + format_prompt`.

## 🚀 Evaluation
Please refer to the evaluation guidelines in our [github repo](https://github.com/InternRobotics/MMSI-Video-Bench).

## 🏆 Leaderboard

<details> <summary>📦 Uniform-50 Setting</summary>

| Model                      | Avg.(%) | Type        |
|----------------------------|---------|-------------|
| Human                   | 96.40   | Baseline    |
|🥇Gemini 3 pro            | 37.97   | Proprietary |
|🥈 O3                      | 36.98   | Proprietary |
|🥉GPT-5                      | 36.80   | Proprietary |
| Gemini 2.5 Flash           | 35.44   | Proprietary |
| Gemini 2.5 Flash (Thinking) | 35.17   | Proprietary |
| Seed-1.6-vision            | 34.87   | Proprietary |
| Claude-haiku-4.5           | 34.27   | Proprietary |
| O4-mini                    | 34.18   | Proprietary |
| QwenVL2.5-72B              | 32.73   | Open-Source |
| InternVL3-78B              | 32.55   | Open-Source |
| Doubao-1.5-thinking        | 31.65   | Proprietary |
| GPT-4o                     | 31.56   | Proprietary |
| InternVL2.5-78B            | 31.37   | Open-Source |
| InternVL2.5-38B            | 31.01   | Open-Source |
| QwenVL3-30B (Thinking)      | 30.83   | Open-Source |
| LLaVA-Video-72B            | 30.38   | Open-Source |
| InternVL3-8B               | 30.38   | Open-Source |
| QwenVL2.5-VL-7B-Instruct   | 29.66   | Open-Source |
| InternVL2.5-8B             | 29.11   | Open-Source |
| InternVL3-38B              | 28.84   | Open-Source |
| QwenVL3-30B                | 28.75   | Open-Source |
| QwenVL2.5-32B              | 28.57   | Open-Source |
| LLaVA-Video-7B             | 28.48   | Open-Source |
| QwenVL3-8B                 | 27.58   | Open-Source |
| InternVideo2.5-8B          | 27.40   | Open-Source |
| Random Guessing            | 24.10   | Baseline    |

</details>

<details> <summary>📦 Sufficient-Coverage Setting</summary>

| Model                      | Avg.(%) | Type        |
|----------------------------|---------|-------------|
| Human                      | 96.4    | Baseline    |
| 🥇O3                         | 37.34   | Proprietary |
| 🥈Gemini 2.5 Flash (Thinking) | 36.71   | Proprietary |
| 🥉Gemini 2.5 Flash           | 36.62   | Proprietary |
| O4-mini                    | 35.08   | Proprietary |
| QwenVL2.5-32B              | 32.37   | Open-Source |
| QwenVL2.5-72B              | 31.83   | Open-Source |
| InternVL3-8B               | 29.57   | Open-Source |
| QwenVL3-30B                | 29.11   | Open-Source |
| QwenVL3-8B                 | 29.09   | Open-Source |
| QwenVL2.5-7B               | 28.84   | Open-Source |
| InternVL2.5-8B             | 28.66   | Open-Source |
| GPT-4o                     | 28.12   | Proprietary |
| QwenVL3-30B (Thinking)      | 28.03   | Open-Source |
| InternVideo2.5-8B          | 26.85   | Open-Source |
| Random Guessing            | 24.10   | Baseline    |

</details>

<details> <summary>🤖 Robot Sub-bench</summary>

| Model                      | Avg.(%) | Type        |
|----------------------------|---------|-------------|
| 🥇Gemini 3 Pro               | 40.20   | Proprietary |
| 🥈Gemini 2.5 Flash (Thinking) | 39.71   | Proprietary |
| 🥉Seed-1.6-vision            | 39.34   | Proprietary |
| O3                         | 39.22   | Proprietary |
| QwenVL2.5-72B              | 37.75   | Open-Source |
| InternVL3-8B               | 37.75   | Open-Source |
| GPT-5                      | 37.75   | Proprietary |
| InternVL2.5-38B            | 36.27   | Open-Source |
| Doubao-1.5-thinking        | 36.07   | Proprietary |
| Gemini 2.5 Flash           | 35.78   | Proprietary |
| O4-mini                    | 35.29   | Proprietary |
| QwenVL2.5-7B               | 34.8    | Open-Source |
| InternVL2.5-78B            | 34.8    | Open-Source |
| Claude-haiku-4.5           | 34.8    | Proprietary |
| InternVL3-78B              | 34.31   | Open-Source |
| LLaVA-Video-72B            | 34.31   | Open-Source |
| QwenVL3-30B                | 32.84   | Open-Source |
| QwenVL2.5-32B              | 32.84   | Open-Source |
| QwenVL3-8B                 | 32.12   | Open-Source |
| InternVideo2.5-8B          | 29.90   | Open-Source |
| GPT-4o                     | 29.90   | Proprietary |
| InternVL2.5-8B             | 28.43   | Open-Source |
| InternVL3-38B              | 27.94   | Open-Source |
| QwenVL3-30B (Thinking)      | 27.94   | Open-Source |
| LLaVA-Video-7B             | 24.51   | Open-Source |


</details>

<details> <summary>🏠 Indoor Scene Perception Sub-bench</summary>

| Model                      | Avg.(%) | Type        |
|----------------------------|---------|-------------|
| 🥇GPT-5                      | 41.68   | Proprietary |
| 🥈O3                         | 40.73   | Proprietary |
| 🥉Gemini 2.5 Flash           | 39.39   | Proprietary |
| Gemini 3 Pro               | 39.39   | Proprietary |
| Gemini 2.5 Flash (Thinking) | 37.86   | Proprietary |
| O4-mini                    | 37.48   | Proprietary |
| Seed-1.6-vision            | 34.2    | Proprietary |
| Claude-haiku-4.5           | 33.46   | Proprietary |
| Doubao-1.5-thinking        | 33.04   | Proprietary |
| InternVL3-78B              | 32.5    | Open-Source |
| QwenVL3-30B (Thinking)      | 32.31   | Open-Source |
| GPT-4o                     | 31.74   | Proprietary |
| QwenVL2.5-72B              | 30.78   | Open-Source |
| InternVL2.5-78B            | 30.4    | Open-Source |
| QwenVL3-30B                | 30.02   | Open-Source |
| QwenVL2.5-32B              | 29.64   | Open-Source |
| InternVL2.5-8B             | 29.45   | Open-Source |
| InternVL3-38B              | 29.06   | Open-Source |
| QwenVL3-8B                 | 28.68   | Open-Source |
| InternVL2.5-38B            | 28.3    | Open-Source |
| LLaVA-Video-72B            | 28.11   | Open-Source |
| InternVL3-8B               | 27.72   | Open-Source |
| LLaVA-Video-7B             | 27.53   | Open-Source |
| QwenVL2.5-7B               | 27.15   | Open-Source |
| InternVideo2.5-8B          | 26.77   | Open-Source |


</details>

<details> <summary>📍 Grounding Sub-bench</summary>

| Model                      | Avg.(%) | Type        |
|----------------------------|---------|-------------|
| 🥇Gemini 2.5 Flash           | 38.81   | Proprietary |
| 🥈Gemini 2.5 Flash (Thinking) | 38.21   | Proprietary |
| 🥉O3                         | 37.61   | Proprietary |
| Doubao-1.5-thinking        | 37.05   | Proprietary |
| InternVL3-78B              | 35.52   | Open-Source |
| GPT-5                      | 35.22   | Proprietary |
| Gemini 3 Pro               | 35.22   | Proprietary |
| O4-mini                    | 34.33   | Proprietary |
| QwenVL2.5-72B              | 34.33   | Open-Source |
| Seed-1.6-vision            | 33.04   | Proprietary |
| Claude-haiku-4.5           | 32.84   | Proprietary |
| InternVL2.5-38B            | 31.94   | Open-Source |
| InternVL3-8B               | 31.94   | Open-Source |
| GPT-4o                     | 31.94   | Proprietary |
| QwenVL3-30B (Thinking)      | 31.64   | Open-Source |
| QwenVL2.5-32B              | 31.04   | Open-Source |
| LLaVA-Video-72B            | 31.04   | Open-Source |
| InternVL3-38B              | 30.45   | Open-Source |
| InternVL2.5-8B             | 30.15   | Open-Source |
| InternVL2.5-78B            | 29.85   | Open-Source |
| QwenVL3-30B                | 29.25   | Open-Source |
| QwenVL2.5-7B               | 28.66   | Open-Source |
| QwenVL3-8B                 | 28.66   | Open-Source |
| InternVideo2.5-8B          | 27.76   | Open-Source |
| LLaVA-Video-7B             | 27.16   | Open-Source |


</details>

*Note: For the three sub-benchmarks, we take the higher score of each model across the two settings for easier presentation.*