OpenVLM SEED-Bench 2: leaderboard
OpenCompass OpenVLM evaluation of SEED-Bench 2: a hierarchical multimodal benchmark testing image, video, and interleaved image-text understanding across 27 dimensions.
Metric: Accuracy (%). Source: huggingface.co. Status: saturated. 59 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4.1 (2025-04-14) | 76 |
| 2 | GPT-4.1 Mini | 75 |
| 3 | Gemini 1.5 Flash | 70.2 |
| 4 | InternVL3-8B | 67.1 |
| 5 | GPT-4.1 Nano | 64.4 |
| 6 | Gemma 3 27B | 59.9 |
| 7 | Pixtral-12B | 57.9 |
| 8 | Gemma 3 12B | 57.6 |
| 9 | Gemma 3 4B | 52.7 |
| 10 | Aquila-VL-2B | 49.3 |
Interactive version: theaggregate.ai/benchmark?slug=openvlm-seed-bench-2 · How It Works · Data refreshed daily, snapshot 2026-09-05.