Atlas

Benchmarks

← All benchmarks

MMLU Pro - Physics

General QA · 2024-06-03

The Physics split of MMLU Pro. This child benchmark separates a source-reported subtask or subtrack from the parent aggregate so scores at different grains do not share one benchmark_slug.

Top models (higher is better)

ModelScore
Claude Fable 5.196.7
Gemini 3.8 Flash94.8
Gemini 3.1 Pro Preview94.7
Gemini 3.7 Flash94.3
Opus 4.694.1
Gemini 3.5 Flash94.1
Muse Spark 1.193.8
Gemini 3 Pro Preview93.7
Opus 4.893.5
Qwen3.7-Max (source unspecified)93.5
DeepSeek-V4-Pro93.5
Gemini 3.6 Flash93.5
Opus 4.793.4
Gemini 3 Flash Preview93.4
Muse Spark 1.293.4
Qwen3.8-Max93.3
GPT-5.6 Sol92.8
Qwen3.5 Plus (2026-02-15)92.8
Qwen3.6 Plus (2026-04-02)92.8
Muse Spark92.7
Sonnet 4.692.6
Kimi K392.6
DeepSeek V4 Pro 081392.5
DeepSeek-V4-Flash-073192.4
GLM-5.192.3
Grok 4.392.3
GLM-5.292.2
Inkling-Small92.2
Nemotron 3 Ultra 550B A55B (NVFP4)92.2
GLM-5.392.1
Grok 4.692.1
Kimi K2.692.1
Gemini 3.1 Flash-Lite Preview92.0
Grok 4.592.0
Inkling92.0
GLM-5.3 Flash91.8
GPT-5.6 Luna91.8
Sonnet 4.591.7
GLM-591.7
Sonnet 591.5
Loading Atlas data…