List your company

Multi-Task on GPQA

41.7Avg@32 AccuracyIBTPO
20.3825.91531.4536.985May 27, 2026

Evaluation results

MethodLinks
IBTPOBase Model=Qwen3-8B-Base2026.0541.7
TreeRLBase Model=Qwen3-8B-Base2026.0539.8
TreePOBase Model=Qwen3-8B-Base2026.0539.2
GRPO w/ Entropy Reg†Base Model=Qwen3-8B-Base2026.0539.1
IBROBase Model=Qwen3-8B-Base2026.0538.5
Vanilla GRPOBase Model=Qwen3-8B-Base2026.0538.1
GRPO w/ Clip-higherBase Model=Qwen3-8B-Base2026.0537.8
IBTPOBase Model=Qwen3-1.7B-Base2026.0529
TreeRLBase Model=Qwen3-1.7B-Base2026.0526.8
Vanilla GRPOBase Model=Qwen3-1.7B-Base2026.0526.5
TreePOBase Model=Qwen3-1.7B-Base2026.0526.2
Initial ModelBase Model=Qwen3-8B-Base2026.0526.1
GRPO w/ Clip-higherBase Model=Qwen3-1.7B-Base2026.0524.8
GRPO w/ Entropy Reg†Base Model=Qwen3-1.7B-Base2026.0524.7
IBROBase Model=Qwen3-1.7B-Base2026.0524.7
Initial ModelBase Model=Qwen3-1.7B-Base2026.0521.2