CoreCognition Single Image Intuitivephysics: leaderboard

CoreCognition Single Image Intuitivephysics — CoreCognition native component accuracy and reported mean. Image+Video, image-only and single-image populations are separate; no pooling or replacement across cohorts. Metric: Accuracy (%). Source: https://grow-ai-like-a-child.github.io/CoreCognition/leaderboard

Metric: Accuracy (%). Source: grow-ai-like-a-child.github.io. Saturation forecast: Estimated already saturated. 33 models tracked. Updated 16 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1qwen2.5-vl-72b-instruct_videoCoreCognition run: native-setting=qwen2.5-vl-72b-instruct_video64.55
2qwen2.5-vl-3b-instruct_videoCoreCognition run: native-setting=qwen2.5-vl-3b-instruct_video61.52
3InternVL2.5-26B60.91#2269
4InternVL2.5-78B59.7#1044
5VideoLLaMA2.1-7B-16F59.09#1924
6qwen2.5-vl-7b-instruct_videoCoreCognition run: native-setting=qwen2.5-vl-7b-instruct_video58.79
7LLaVA-Video-72B-Qwen257.88#1228
8InternVL2.5-4B57.27#2583
9Llama 3.2 11B Instruct56.97#5235
10llava-video-7B-qwen2-video-only56.36
11InternVL2.5-2B54.39#6932
12VideoLLaMA2-72B53.33#1355
13InternVL2.5-38B52.58#1413
14Llama 3.2 90B Vision Instruct51.97#1769
15chat-univi51.97

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=corecognition-single-image-intuitivephysics · How It Works · Data refreshed daily, snapshot 2026-11-16.