Mathematical Reasoning on AIME 2025 (Accuracy, Avg, Speedup)
99.2AccuracyTRICE-30B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TRICE-30BTool Usage=true, Parameter Scale=~30B, Evaluation Protocol=unified2026.05 | 99.2 | — | — | |
| GLM-4.7-Flash w/ recipeTool Usage=true, Parameter Scale=~30B, Evaluation Protocol=unified2026.05 | 98.3 | — | — | |
| TRICE-4BTool Usage=true, Parameter Scale=<10B, Evaluation Protocol=unified2026.05 | 96.7 | — | — | |
| Nemotron-3-Nano-30B-A3BTool Usage=true, Parameter Scale=~30B, Evaluation Protocol=unified2026.05 | 96.7 | — | — | |
| GLM-5Params=744B2026.06 | 96.7 | — | — | |
| VibeThinker-3B + CLRParams=3B, Post-training optimization=CLR2026.06 | 96.7 | — | — | |
| Kimi K2.5Params=1T2026.06 | 96.1 | — | — | |
| Gemini 3 ProParams=N/A2026.06 | 96 | — | — | |
| GLM-4.7-FlashTool Usage=true, Parameter Scale=~30B, Evaluation Protocol=unified2026.05 | 95 | — | — | |
| GPT-5 (high)Params=N/A2026.06 | 94.6 | — | — | |
| Qwen3.5-35B-A3BTool Usage=false, Parameter Scale=~30B, Evaluation Protocol=unified2026.05 | 94.2 | — | — | |
| MiMo v2 FlashParams=309B2026.06 | 94.1 | — | — | |
| Qwen3.6 PlusParams=N/A2026.06 | 93.3 | — | — | |
| DeepSeek V3.2Params=671B2026.06 | 93.1 | — | — | |
| Claude Opus 4.5Params=N/A2026.06 | 92.8 | — | — | |
| GPT-OSS (high)Params=120B2026.06 | 92.5 | — | — | |
| Grok 4Params=N/A2026.06 | 91.7 | — | — | |
| VibeThinker-3BParams=3B2026.06 | 91.4 | — | — | |
| ASTER-4B†Tool Usage=true, Parameter Scale=<10B, Evaluation Protocol=original2026.05 | 90 | — | — | |
| VeryTraceModel=Q3-80b-T2026.06 | 90 | — | — | |
| TRICE-30BTool Usage=false, Parameter Scale=~30B, Evaluation Protocol=unified2026.05 | 89.2 | — | — | |
| OpenAI o3 (high)Params=N/A2026.06 | 88.9 | — | — | |
| Qwen3-30B-A3B-Thinking-2507Tool Usage=false, Parameter Scale=~30B, Evaluation Protocol=unified2026.05 | 88.8 | — | — | |
| DeepSeek R1 0528Params=671B2026.06 | 87.5 | — | — | |
| GPT-OSS-20BTool Usage=true, Parameter Scale=~30B, Evaluation Protocol=unified2026.05 | 86.7 | — | — | |
| Gemini 2.5 ProParams=N/A2026.06 | 86.7 | — | — | |
| CoVeModel=Q3-80b-T2026.06 | 86.67 | — | — | |
| AgentMath-30B-A3B†Tool Usage=true, Parameter Scale=~30B, Evaluation Protocol=original2026.05 | 86.4 | — | — | |
| Qwen3.5-9BTool Usage=false, Parameter Scale=<10B, Evaluation Protocol=unified2026.05 | 85.8 | — | — | |
| AgentMath-8B†Tool Usage=true, Parameter Scale=<10B, Evaluation Protocol=original2026.05 | 84.7 | — | — | |
| VanillaModel=Q3-80b-T2026.06 | 83.33 | — | — | |
| NPModel=Q3-80b-T2026.06 | 83.33 | — | — | |
| Qwen3-4B-Thinking-2507Tool Usage=false, Parameter Scale=<10B, Evaluation Protocol=unified2026.05 | 82.5 | — | — | |
| VeryTraceModel=Q3-80b-I2026.06 | 80 | — | — | |
| TRICE-4BTool Usage=false, Parameter Scale=<10B, Evaluation Protocol=unified2026.05 | 79.2 | — | — | |
| Qwen3.5-4BTool Usage=false, Parameter Scale=<10B, Evaluation Protocol=unified2026.05 | 75.8 | — | — | |
| SeirênesBackbone=Qwen3-30B-A3B, same budget=true2026.05 | 74.1 | — | — | |
| CoVeModel=Q3-80b-I2026.06 | 73.33 | — | — | |
| NPModel=Q3-80b-I2026.06 | 73.33 | — | — | |
| Gemini 2.5 FlashParams=N/A2026.06 | 72 | — | — | |
| Glob. non-uniBackbone Model=Qwen3-8B2025.10 | 70 | — | — | |
| Loc. uniBackbone Model=Qwen3-8B2025.10 | 69 | — | — | |
| Qwen3-30B-A3B-Instruct-2507Tool Usage=false, Parameter Scale=~30B, Evaluation Protocol=unified2026.05 | 67.5 | — | — | |
| Qwen3-4B-LambdaGRPOModel Architecture=Qwen3, Model Size=4B, Training Algorithm=LambdaGRPO2026.05 | 67.08 | 76.49 | — | |
| LamPOBackbone=Qwen3-4B2026.05 | 67.08 | — | — | |
| Mean Acc.Backbone Model=Qwen3-8B2025.10 | 67 | — | — | |
| DAPOBackbone=Qwen3-30B-A3B, same budget=true2026.05 | 66.8 | — | — | |
| ReLATModel=Qwen3-14B2026.06 | 66.7 | — | — | |
| Glob. uniBackbone Model=Qwen3-8B2025.10 | 66 | — | — | |
| Qwen3-4B-GRPOModel Architecture=Qwen3, Model Size=4B, Training Algorithm=GRPO2026.05 | 65.83 | 75.4 | — | |
| GRPOBackbone=Qwen3-4B2026.05 | 65.83 | — | — | |
| SeirênesBackbone=Qwen3-4B-Instruct, same budget=true2026.05 | 64.9 | — | — | |
| Dr.GRPOBackbone=Qwen3-4B-Instruct, same budget=true2026.05 | 63.3 | — | — | |
| ReLATModel=Qwen3-8B2026.06 | 63.3 | — | — | |
| LatentMASModel=Qwen3-14B2026.06 | 63.3 | — | — | |
| Self-Cert.Backbone Model=Qwen3-8B2025.10 | 63 | — | — | |
| Loc. non-uniBackbone Model=Qwen3-8B2025.10 | 63 | — | — | |
| Qwen3-30B-A3BBackbone=Qwen3-30B-A3B2026.05 | 60.6 | — | — | |
| DAPOBackbone=Qwen3-4B-Instruct, same budget=true2026.05 | 60.2 | — | — | |
| High Conf.Backbone Model=Qwen3-8B2025.10 | 60 | — | — | |
| Low Ent.Backbone Model=Qwen3-8B2025.10 | 60 | — | — | |
| Qwen3-4BModel Architecture=Qwen3, Model Size=4B2026.05 | 60 | 73.18 | — | |
| CoTBackbone=Qwen3-4B2026.05 | 60 | — | — | |
| TextMASModel=Qwen3-14B2026.06 | 60 | — | — | |
| InTBackbone=Qwen3-4B-Instruct2026.05 | 58.9 | — | — | |
| ReLATModel=Qwen3-4B2026.06 | 56.7 | — | — | |
| SingleModel=Qwen3-14B2026.06 | 56.7 | — | — | |
| VanillaModel=Q3-80b-I2026.06 | 53.33 | — | — | |
| TextMASModel=Qwen3-8B2026.06 | 53.3 | — | — | |
| LatentMASModel=Qwen3-8B2026.06 | 53.3 | — | — | |
| Loc. uniBackbone Model=DS-R1-Distill-Qwen-7B2025.10 | 53 | — | — | |
| Glob. non-uniBackbone Model=DS-R1-Distill-Qwen-7B2025.10 | 52 | — | — | |
| LatentMASModel=Qwen3-4B2026.06 | 50 | — | — | |
| Scaf-GRPOBackbone=Qwen3-4B-Instruct2026.05 | 49.1 | — | — | |
| SAGEBackbone=Qwen3-4B-Instruct2026.05 | 49.1 | — | — | |
| Qwen3-4B-InstructBackbone=Qwen3-4B-Instruct2026.05 | 48.8 | — | — | |
| Self-Cert.Backbone Model=DS-R1-Distill-Qwen-7B2025.10 | 48 | — | — | |
| High Conf.Backbone Model=DS-R1-Distill-Qwen-7B2025.10 | 48 | — | — | |
| Low Ent.Backbone Model=DS-R1-Distill-Qwen-7B2025.10 | 48 | — | — | |
| SingleModel=Qwen3-8B2026.06 | 46.7 | — | — | |
| GRPO + ARROLBackbone=Qwen-3-4B-Base, Method=GRPO + ARROL2026.03 | 46.67 | 53.54 | 1.63 | |
| GRPO + ARROLBackbone=Qwen-3-8B-Base, Method=GRPO + ARROL2026.03 | 46.67 | 59.53 | 1.62 | |
| PieceHintBase model=DeepSeek-R1-Distill-7B2026.04 | 43.9 | — | — | |
| SCOPE-RLBase Model=Qwen3-4B2025.10 | 43.7 | — | — | |
| SingleModel=Qwen3-4B2026.06 | 43.3 | — | — | |
| TextMASModel=Qwen3-4B2026.06 | 43.3 | — | — | |
| Entropy-RegBase Model=Qwen3-4B, λ=0.012025.10 | 42.6 | — | — | |
| CISPOBase Model=Qwen3-4B2025.10 | 42.6 | — | — | |
| KL-covBase Model=Qwen3-4B2025.10 | 42.3 | — | — | |
| DAPOBase Model=Qwen3-4B2025.10 | 41.7 | — | — | |
| Entropy-AdvBase Model=Qwen3-4B2025.10 | 41.7 | — | — | |
| GRPOBase Model=Qwen3-4B2025.10 | 41.5 | — | — | |
| STARE-O1Scenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.06 | 41.5 | — | — | |
| STARE-C2Scenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.06 | 40.5 | — | — | |
| GRPOBackbone=Qwen-3-4B-Base, Method=GRPO2026.03 | 40 | 51.01 | — | |
| GRPOBackbone=Qwen-3-8B-Base, Method=GRPO2026.03 | 40 | 56.66 | — | |
| Mean Acc.Backbone Model=DS-R1-Distill-Qwen-7B2025.10 | 40 | — | — | |
| ReLATModel=DeepSeek-R1-Distill-Qwen-7B2026.06 | 40 | — | — | |
| Loc. uniBackbone Model=DS-R1-Distill-Llama-8B2025.10 | 39 | — | — | |
| JustRLScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.06 | 38.8 | — | — |