List your company
ResearchFormerly WizWand

General Reasoning on GPQA Diamond (ACC, A.Tok, epsilon^3)

72.6AccuracyQwenLong L1.5-30B
24.936837.310949.68562.0591May 10, 2026May 14, 2026May 19, 2026May 24, 2026May 29, 2026Jun 3, 2026Jun 8, 2026
Updated 4mo ago

Evaluation results

MethodLinks
QwenLong L1.5-30BModel=QwenLong L1.5-30B2026.0572.6——
GoLongRL (w. GRPO)Model=Qwen3-30B-A3B-Thinking-2507, Optimization Method=GRPO2026.0572.3——
Qwen3-30B-A3B-Thinking-2507 BaseModel=Qwen3-30B-A3B-Thinking-2507, Optimization Method=Base2026.0570.1——
GoLongRL (w. TMN-Reweight)Model=Qwen3-4B-Thinking-2507, Optimization Method=TMN-Reweight2026.0567.7——
Qwen3-4B-Thinking-2507 BaseModel=Qwen3-4B-Thinking-2507, Optimization Method=Base2026.0565.4——
AdaptThinkBackbone=DeepSeek-R1-Distill-Qwen-7B, Trained with same data and RL paradigm=True2026.0548.987,124.4434
R-FewModel Backbone=Qwen3-8B-Base2026.0646.5——
INFUSERModel Backbone=Qwen3-8B-Base2026.0645.48——
AZRModel Backbone=Qwen3-8B-Base2026.0644.14——
R-ZeroModel Backbone=Qwen3-8B-Base2026.0642.73——
R-FewModel Backbone=Qwen3-4B-Base2026.0639.9——
SPICEModel Backbone=Qwen3-4B-Base2026.0639.4——
SPICEModel Backbone=Qwen3-8B-Base2026.0639.4——
VPG-EABackbone=DeepSeek-R1-Distill-Qwen-7B, Trained with same data and RL paradigm=True2026.0538.883,892.6239
AZRModel Backbone=Qwen3-4B-Base2026.0637.17——
JETBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0536.892,915.7947
BaseModel Backbone=Qwen3-8B-Base2026.0636.87——
Original ModelBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0536.866,471.2621
R1-VeriThinkerBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0536.864,060.4333
INFUSERModel Backbone=Qwen3-4B-Base2026.0636.8——
AutoThinkBackbone=DeepSeek-R1-Distill-Qwen-7B, Trained with same data and RL paradigm=True2026.0534.854,763.0425
R-ZeroModel Backbone=Qwen3-4B-Base2026.0634.44——
LaserBackbone=DeepSeek-R1-Distill-Qwen-7B, Trained with same data and RL paradigm=True2026.0533.833,603.6632
ThinklessBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0532.827,465.2214
BaseModel Backbone=Qwen3-4B-Base2026.0631.41——
LaserBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Trained with same data and RL paradigm=True2026.0530.86,695.5714
VPG-EABackbone=DeepSeek-R1-Distill-Qwen-1.5B, Trained with same data and RL paradigm=True2026.0530.86,237.9815
Original ModelBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0529.797,419.912
AdaptThinkBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Trained with same data and RL paradigm=True2026.0528.75,106.916
JETBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0526.776,140.9812
AutoThinkBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Trained with same data and RL paradigm=True2026.0526.775,768.0812