Benchgen

AI2D — Results

RankModelScore
1north-micro-vision-instruct77.5
2claude-3-5-sonnet0.947
3qwen3-6-plus0.944
4gpt-4o0.942
5qwen3-5-122b-a10b0.933
6mistral-small-3-2-24b-instruct0.929
7qwen3-5-27b0.929
8qwen3-6-35b-a3b0.927
9qwen3-5-35b-a3b0.926
10llama-3-2-90b-instruct0.923
11qwen3-vl-235b-a22b-instruct0.897
12qwen3-vl-32b-instruct0.895
13qwen3-vl-235b-a22b-thinking0.892
14qwen3-vl-32b-thinking0.889
15qwen3-vl-30b-a3b-thinking0.869
16qwen3-vl-8b-instruct0.857
17qwen3-vl-30b-a3b-instruct0.85
18qwen3-vl-4b-thinking0.849
19qwen3-vl-8b-thinking0.849
20gemma-3-27b0.845
21gemma-3-12b0.842
22qwen3-vl-4b-instruct0.841
23qwen2-5-omni-7b0.832
24gemma-3-4b0.748

AI2D — Phases

1 phaseActive

Phase 1·

AI2D

Aug 3, 2026
Current