OpenVLM SEED-Bench 2 Plus - Webs: leaderboard
SEED-Bench-2-Plus Webs accuracy from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): multiple-choice questions on text-rich website screenshots, one of the benchmark's three image types (charts, maps and webs). Accuracy in percent, higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 211 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 (2025-04-14) | 85.9 | #257 |
| 2 | Qwen 2 VL 72B | 85.8 | #450 |
| 3 | Qwen 2.5 VL 7B | 85.3 | #641 |
| 4 | InternVL3-38B | 85 | #413 |
| 5 | InternVL3-78B | 84.8 | #361 |
| 6 | GPT-4.1 Mini | 84.5 | #363 |
| 7 | Gemini 1.5 Pro | 84.5 | #454 |
| 8 | Claude 3.5 Sonnet | 84.2 | #362 |
| 9 | Llama 3.2 90B Vision Instruct | 84.1 | #621 |
| 10 | InternVL3-14B | 83.2 | #506 |
| 11 | InternVL3-8B | 82.6 | #624 |
| 12 | Qwen 2 VL 7B | 82.6 | #743 |
| 13 | Llama 3.2 11B Instruct | 82.4 | #1108 |
| 14 | InternVL2-8B | 82 | #813 |
| 15 | Gemma 3 27B | 80.8 | #606 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-seed-bench-2-plus-webs · How It Works · Data refreshed daily, snapshot 2026-11-05.