Share your thoughts, 1 month free Claude Pro on usSee more

Science on MMLU-Pro

90.1Accuracy

Gemini-3-pro High

Updated 23d ago

Evaluation Results

Method	Links
Gemini-3-pro High 2026.06		90.1
Claude-Opus-4.5 2026.06		89.3
Claude-Sonnet-4.5 2026.06		88
Seed2.0 Pro 2026.06		87
GPT-5.2 High 2026.06		85.9
Base 2026.05		41.9
Base 2026.05		41.9
Self-SFT 2026.05		38.6
Low-SFT 2026.05		37.8
Iter-SFT 2026.05		37.7
Anchored Learning 2026.05		37.7
Anchored Learning 2026.05		37.7
Anchored Learning 2026.05		36.8
Low-SFT 2026.05		36.8
Anchored Learning 2026.05		36.6
Low-SFT 2026.05		36.3
KL-SFT 2026.05		35.5
Base 2026.05		34.7
Base 2026.05		34.7
Base 2026.05		34.7
DFT 2026.05		34.7
Anchored Learning 2026.05		34.3
Anchored Learning 2026.05		33.3
Low-SFT 2026.05		33.3
STM 2026.05		33.3
Low-SFT 2026.05		33.3
Low-SFT 2026.05		33.2
SFT 2026.05		32.7
Iter-SFT 2026.05		32.2
STM 2026.05		32.1
Self-SFT 2026.05		31.2
Iter-SFT 2026.05		30.9
Anchored Learning 2026.05		30.8
Low-SFT 2026.05		29.9
Self-SFT 2026.05		29.6
STM 2026.05		29.2
DFT 2026.05		28
Iter-SFT 2026.05		28
Self-SFT 2026.05		26.9
STM 2026.05		24.3
DFT 2026.05		16.3
KL-SFT 2026.05		10.2
KL-SFT 0.2 2026.05		9.6
DFT 2026.05		8.2
SFT 2026.05		3.1
KL-SFT 2026.05		2
SFT 2026.05		1.2
SFT 2026.05		1.1