Reasoning on HMMT25
82.8Avg@32EqLen-GRPO
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| EqLen-GRPOBackbone=Qwen3-Next2026.04 | 82.8 | — | — | — | — | — | — | — | — | |
| LUSPOBackbone=Qwen3-Next2026.04 | 77.1 | — | — | — | — | — | — | — | — | |
| DAPOBackbone=Qwen3-Next2026.04 | 76.8 | — | — | — | — | — | — | — | — | |
| Dr. GRPOBackbone=Qwen3-Next2026.04 | 76.2 | — | — | — | — | — | — | — | — | |
| GSPOBackbone=Qwen3-Next2026.04 | 75.4 | — | — | — | — | — | — | — | — | |
| GRPOBackbone=Qwen3-Next2026.04 | 74.8 | — | — | — | — | — | — | — | — | |
| BaseBackbone=Qwen3-Next2026.04 | 73.9 | — | — | — | — | — | — | — | — | |
| EqLen-GRPOBackbone=Qwen3-32B2026.04 | 64 | — | — | — | — | — | — | — | — | |
| LUSPOBackbone=Qwen3-32B2026.04 | 58.2 | — | — | — | — | — | — | — | — | |
| DAPOBackbone=Qwen3-32B2026.04 | 57.8 | — | — | — | — | — | — | — | — | |
| Dr. GRPOBackbone=Qwen3-32B2026.04 | 56.1 | — | — | — | — | — | — | — | — | |
| GSPOBackbone=Qwen3-32B2026.04 | 55.4 | — | — | — | — | — | — | — | — | |
| GRPOBackbone=Qwen3-32B2026.04 | 54.6 | — | — | — | — | — | — | — | — | |
| Type IV (STOP)Model=DS-Qwen-3-8B2026.04 | 54.58 | — | — | — | — | — | 311.7 | -80.53 | — | |
| Type IV (STOP)Model=GPT-OSS-20B2026.04 | 52.92 | — | — | — | — | — | 216.1 | -76.27 | — | |
| BaseBackbone=Qwen3-32B2026.04 | 51.5 | — | — | — | — | — | — | — | — | |
| Type IIIModel=DS-Qwen-3-8B2026.04 | 51.25 | — | — | — | — | — | 314 | -80.38 | — | |
| No pruning (Baseline)Model=DS-Qwen-3-8B2026.04 | 51.04 | — | — | — | — | — | 1,601 | — | — | |
| Type IIModel=DS-Qwen-3-8B2026.04 | 50.83 | — | — | — | — | — | 427.8 | -73.28 | — | |
| Type IModel=DS-Qwen-3-8B2026.04 | 50 | — | — | — | — | — | 322.1 | -79.88 | — | |
| No pruning (Baseline)Model=GPT-OSS-20B2026.04 | 48.13 | — | — | — | — | — | 910.8 | — | — | |
| Type IIIModel=GPT-OSS-20B2026.04 | 46.25 | — | — | — | — | — | 228 | -74.97 | — | |
| Type IModel=GPT-OSS-20B2026.04 | 45.83 | — | — | — | — | — | 236.3 | -74.06 | — | |
| Type IIModel=GPT-OSS-20B2026.04 | 45.42 | — | — | — | — | — | 336.9 | -63.01 | — | |
| Type IV (STOP)Model=DS-Qwen-2.5-7B2026.04 | 27.08 | — | — | — | — | — | 211.6 | -73.84 | — | |
| Type IModel=DS-Qwen-2.5-7B2026.04 | 25 | — | — | — | — | — | 219.5 | -72.87 | — | |
| Type IIIModel=DS-Qwen-2.5-7B2026.04 | 24.58 | — | — | — | — | — | 216.1 | -73.28 | — | |
| No pruning (Baseline)Model=DS-Qwen-2.5-7B2026.04 | 23.91 | — | — | — | — | — | 808.9 | — | — | |
| Type IIModel=DS-Qwen-2.5-7B2026.04 | 23.33 | — | — | — | — | — | 330.8 | -59.11 | — | |
| Type IV (STOP)Model=DS-Qwen-2.5-1.5B2026.04 | 17.92 | — | — | — | — | — | 215.5 | -74.84 | — | |
| Type IIIModel=DS-Qwen-2.5-1.5B2026.04 | 17.08 | — | — | — | — | — | 220.9 | -74.21 | — | |
| Type IModel=DS-Qwen-2.5-1.5B2026.04 | 15.83 | — | — | — | — | — | 224.2 | -73.82 | — | |
| No pruning (Baseline)Model=DS-Qwen-2.5-1.5B2026.04 | 15.05 | — | — | — | — | — | 856.4 | — | — | |
| Type IIModel=DS-Qwen-2.5-1.5B2026.04 | 15 | — | — | — | — | — | 337.2 | -60.63 | — | |
| CraEGTemperature=1, Top-p=1, Backbone=Qwen3-1.7B2026.01 | 14.58 | 30.58 | 51.93 | 15.46 | — | — | — | — | — | |
| Standard SamplingTemperature=1, Top-p=1, Backbone=Qwen3-1.7B2026.01 | 13.85 | 27.74 | 51.01 | 14.78 | — | — | — | — | — | |
| CraEGTemperature=0.7, Top-p=0.95, Backbone=Qwen3-1.7B2026.01 | 12.6 | 24.43 | 42.28 | 12.1 | — | — | — | — | — | |
| Standard SamplingTemperature=0.7, Top-p=0.95, Backbone=Qwen3-1.7B2026.01 | 11.88 | 21.44 | 41.53 | 12.02 | — | — | — | — | — | |
| ASCType=Handcrafted, Base Model=DeepSeek-R1-Distill-Llama-8B2026.05 | — | — | — | — | — | — | 582,700 | — | 26.5 | |
| AutoTTSType=Discovered, Base Model=DeepSeek-R1-Distill-Llama-8B, β=12026.05 | — | — | — | — | — | — | 533,900 | — | 27.2 | |
| AutoTTSType=Discovered, Base Model=DeepSeek-R1-Distill-Llama-8B, β=0.52026.05 | — | — | — | — | — | — | 279,000 | — | 26.3 | |
| ESCType=Handcrafted, Base Model=DeepSeek-R1-Distill-Llama-8B2026.05 | — | — | — | — | — | — | 952,100 | — | 26.7 | |
| MULTIVERSEReasoning Type=Parallel2025.12 | — | — | — | — | 1,465.1 | 2.1 | — | — | — | |
| NPR-Inst.Reasoning Type=Parallel2025.12 | — | — | — | — | 2,784.1 | 4.1 | — | — | — | |
| SC@64Type=Handcrafted, Base Model=DeepSeek-R1-Distill-Llama-8B2026.05 | — | — | — | — | — | — | 985,700 | — | 26.7 | |
| SRReasoning Type=Sequential baseline2025.12 | — | — | — | — | 683.8 | 1 | — | — | — |