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

Mathematical Reasoning on HMMT25 (Avg@16 accuracy (%))

33.2Avg@16 Accuracy

TRD

Updated 1mo ago

Evaluation Results

Method	Links
TRD 2026.06		33.2
Base 2026.06		31.3
Reverse KL 2026.06		31.3
Reverse KL w/ Top-K 2026.06		30.6
Forward KL 2026.06		30.4
Forward KL w/ Clip 2026.06		29.8
DASD 2026.05		25
DASD 2026.05		25
TRD 2026.06		24.4
Base 2026.06		24.2
Reverse KL w/ Top-K 2026.06		23.8
Forward KL 2026.06		23.3
Forward KL w/ Clip 2026.06		22.9
Reverse KL 2026.06		22.9
GRPO 2026.05		22.1
HEPO 2026.05		21.2
GRPO 2026.05		20.8
RLSD 2026.05		20.8
RLSD 2026.05		20.4
HEPO 2026.05		20.4
DASD 2026.05		17.7
SRPO 2026.05		17.7
SRPO 2026.05		17.5
HEPO 2026.05		15.6
RLSD 2026.05		15.2
Qwen3-4B 2026.05		15.2
OPSD 2026.05		15
SDPO 2026.05		14.4
SDPO 2026.05		13.3
Qwen3-8B 2026.05		13.1
OPSD 2026.05		12.9
GRPO 2026.05		11.5
SRPO 2026.05		10.2
OPSD 2026.05		8.5
SDPO 2026.05		8.1
Qwen3-1.7B 2026.05		7.9