Atlas

Benchmarks

← All benchmarks

MindCube Tiny

Multimodal · 2025-06-26

MindCube Tiny is a compact split of MindCube, which tests spatial mental modeling from limited visual observations. Tasks cover position and cognitive mapping, orientation and perspective taking, and dynamics and mental simulation.

Top models (higher is better)

ModelScore
Gemini 4 Argon96.7
Claude Opus 5.593.1
GPT-6 Astra92.7
Gemini 3.8 Flash92.2
Muse Wonder91.1
GPT-6.1 Sol89.1
Gemini 3.7 Flash88.4
Claude Fable 5.184.5
Claude Opus 584.3
Gemini 3.1 Pro Preview84.1
Gemini 3.5 Flash83.9
Kimi K383.3
Claude Fable 583.1
Muse Spark Open81.3
Muse Spark 1.180.1
GPT-5.579.9
Gemini 3 Flash Preview78.3
Grok 4.578.1
Gemini 3 Pro Preview77.3
GPT-5.6 Sol76.8
Kimi K2.675.6
Grok 4.372.2
GPT-5.470.4
Grok 4.2067.7
Opus 4.667.0
Opus 4.864.9
Grok 464.7
Opus 4.763.9
GPT-5.6 Terra62.6
GPT-5.162.0
GPT-5.261.7
Opus 4.561.2
GPT-5 Mini61.1
Sonnet 4.660.0
GPT-559.7
Gemini 2.5 Pro59.6
Sonnet 4.558.3
GPT-5.6 Luna57.8
o357.6
GPT-5.4 Mini56.5
Loading Atlas data…