Mathematical Reasoning on GSM8K (pass@1, Avg Rank)
92.87Pass@1TYLER Stage 2
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TYLER Stage 2Backbone=Qwen3-4B2026.06 | 92.87 | — | |
| TYLER Stage 1Backbone=Qwen3-4B2026.06 | 91.66 | — | |
| Soft-ThinkingBackbone=Qwen3-4B2026.06 | 91.51 | — | |
| CoTBackbone=Qwen3-4B2026.06 | 90.52 | — | |
| SwiReasoningBackbone=Qwen3-4B2026.06 | 90.21 | — | |
| GRPOBackbone=Qwen3-4B2026.06 | 90.14 | — | |
| MemGenBackbone=Qwen3-4B2026.06 | 89.65 | — | |
| GRPO2026.04 | 88.7 | 2.43 | |
| SoftCoTBackbone=Qwen3-4B2026.06 | 88.32 | — | |
| SFTBackbone=Qwen3-4B2026.06 | 86.96 | — | |
| Reinforce++2026.04 | 86.58 | 4.57 | |
| SEARL2026.04 | 86.2 | 1.43 | |
| distillm22026.04 | 85.7 | — | |
| TYLER Stage 2Backbone=SmolLM3-3B2026.06 | 85.21 | — | |
| VCRD-Probapproach=PRM-free2026.04 | 84.8 | — | |
| MemGenBackbone=SmolLM3-3B2026.06 | 84.15 | — | |
| SoftCoTBackbone=SmolLM3-3B2026.06 | 83.85 | — | |
| TYLER Stage 1Backbone=SmolLM3-3B2026.06 | 83.24 | — | |
| SFTBackbone=SmolLM3-3B2026.06 | 83.02 | — | |
| ARPO2026.04 | 82.41 | 3.57 | |
| DAPO2026.04 | 80.59 | 3 | |
| Soft-ThinkingBackbone=SmolLM3-3B2026.06 | 76.65 | — | |
| SwiReasoningBackbone=SmolLM3-3B2026.06 | 74.98 | — | |
| GRPOBackbone=SmolLM3-3B2026.06 | 72.71 | — | |
| Qwen3-8B-BaseSeed=Avg2025.11 | 70.93 | — | |
| P-POTS+MirrorBackbone=LLaDA-8B-Instruct, Seed=Avg2025.11 | 60.53 | — | |
| CoTBackbone=SmolLM3-3B2026.06 | 58.91 | — | |
| ClippedBackbone=LLaDA-8B-Instruct, Seed=Avg2025.11 | 56.18 | — | |
| MIRRORBackbone=LLaDA-8B-Instruct, Seed=Avg2025.11 | 53.7 | — | |
| TIR Prompt2026.04 | 22.59 | 5.29 |