SEEDBench IMG: leaderboard
19K multiple-choice questions with human annotations evaluating multimodal LLMs on image understanding across 9 dimensions including scene and spatial comprehension.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 226 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3-78B | 78.7 | #352 |
| 2 | InternVL3-38B | 78.3 | #405 |
| 3 | GPT-4.1 (2025-04-14) | 78 | #250 |
| 4 | Qwen 2 VL 72B | 77.9 | #460 |
| 5 | GPT-4.1 Mini | 77.6 | #355 |
| 6 | InternVL3-14B | 77.5 | #505 |
| 7 | InternVL3-8B | 77.1 | #620 |
| 8 | Qwen 2.5 VL 7B | 77 | #643 |
| 9 | Llama 3.2 90B Vision Instruct | 76.8 | #622 |
| 10 | Grok 2 (1212) | 76.3 | #496 |
| 11 | Qwen 2 VL 7B | 76 | #750 |
| 12 | Gemini 1.5 Pro | 76 | #445 |
| 13 | InternVL2-8B | 75.4 | #817 |
| 14 | Moonshot v1 8K | 75.2 | #617 |
| 15 | Gemini 1.5 Flash | 75 | #618 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seedbench-img · How It Works · Data refreshed daily, snapshot 2026-10-20.