Atlas

Benchmarks

← All benchmarks

IntPhys 2

Multimodal · 2025-06-11

IntPhys 2 evaluates intuitive physics understanding with video clips that test object permanence, solidity, and spatio-temporal continuity. Models identify physically plausible versus impossible physical-world scenes.

Top models (higher is better)

ModelScore
GPT-6 Astra77.4
Gemini 4 Argon74.6
GPT-6.1 Sol72.3
Muse Wonder70.2
Claude Opus 5.568.3
Gemini 3.7 Flash68.2
Claude Fable 5.167.7
Claude Fable 567.1
Gemini 3.5 Flash67.0
Muse Spark 1.166.8
Claude Opus 565.1
Muse Spark Open64.9
GPT-5.6 Sol63.9
Gemini 3 Flash Preview63.4
Grok 4.2062.6
Gemini 3.8 Flash62.5
Grok 4.561.4
GPT-5.6 Terra61.3
Kimi K2.660.0
GPT-5.559.9
GPT-5 Mini59.2
Kimi K358.6
Gemini 3.1 Flash-Lite Preview58.4
GPT-5.258.3
GPT-5.6 Luna58.0
Gemini 2.5 Flash58.0
Gemini 3 Pro Preview56.9
Grok 4.356.9
GPT-5.456.4
Opus 4.556.3
GPT-556.1
Opus 4.756.0
Gemini 2.5 Pro56.0
Opus 4.855.1
Grok 4.1 Fast55.0
Grok 454.9
o354.7
Sonnet 4.554.4
Gemini 2.5 Flash-Lite54.1
Opus 4.653.6
Loading Atlas data…