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

Science Reasoning on GPQA (Accuracy)

59.22Accuracy (GPQA)

GSPO + NSR

Updated 5d ago

Evaluation Results

Method	Links
GSPO + NSR 2026.05		59.22
GSPO 2026.05		58.96
In-place feedback 2025.10		58.7
In-place feedback 2025.10		53.2
DAPO + NSR 2026.05		53.16
DAPO 2026.05		52.02
Multi-turn 2025.10		51.6
Multi-turn 2025.10		50.8
Memory 2025.10		50.8
SoftThink 2025.11		50.3
Ouro 2025.11		50
In-place feedback 2025.10		49.2
TaH+ 2025.11		49
TaH 2025.11		48.5
Multi-turn 2025.10		48.4
Memory 2025.10		47.6
In-place feedback 2025.10		46.8
IBTPO 2026.05		46.6
Memory 2025.10		46.6
Standard 2025.11		46.2
Multi-turn 2025.10		46
IBRO 2026.05		44.7
Memory 2025.10		44.4
GRPO w/ Entropy Reg 2026.05		44.2
TreeRL 2026.05		44.2
Vanilla GRPO 2026.05		43.7
In-place feedback 2025.10		43.7
DAPO + NSR 2026.05		42.99
In-place feedback 2025.10		42.9
Multi-turn 2025.10		42.9
Multi-turn 2025.10		42.1
Memory 2025.10		40.5
Memory 2025.10		39.7
TaH+ 2025.11		39.4
DAPO 2026.05		39.27
Initial Model 2026.05		38.6
Ouro 2025.11		35.4
In-place feedback 2025.10		34.9
Qwen3.5 2B 2026.05		34.3
Multi-turn 2025.10		34.1
MobileMoE-L 2026.05		33.8
SoftThink 2025.11		33.3
TaH 2025.11		33.3
Memory 2025.10		33.3
Ouro 2025.11		32.8
In-place feedback 2025.10		31.8
TaH+ 2025.11		31.3
Standard 2025.11		31.1
AlwaysThink 2025.11		30.8
AlwaysThink 2025.11		30.5
Standard 2025.11		30.3
Memory 2025.10		30.2
OLMo 2 1B 2026.05		29.8
SmolLM2 1.7B 2026.05		29.8
Multi-turn 2025.10		29.4
TaH 2025.11		29
Llama 3.2 1B 2026.05		28.8
MobileMoE-S 2026.05		27.8
Qwen3.5 0.8B 2026.05		26.3
Gemma 3 270M 2026.05		25.8
SmolLM2 360M 2026.05		25.8
Gemma 3 1B 2026.05		25.3
MobileMoE-M 2026.05		24.8
SoftThink 2025.11		24.7
OLMoE-1B-7B 2026.05		24.2
MobileLLM-Pro 2026.05		23.2
Multi-turn 2025.10		18.3
Memory 2025.10		18.3
In-place feedback 2025.10		18.3
Multi-turn 2025.10		17.5
Memory 2025.10		17.5
In-place feedback 2025.10		17.5