Code Generation on LiveCodeBench (Accuracy)
73.2AccuracyTRICE-30B
Evaluation Results
| Method | Links | |
|---|---|---|
| TRICE-30BUse Tool=true2026.05 | 73.2 | |
| CyclicReflexBackbone=Qwen3-14B, Decoding Strategy=CyclicReflex2025.06 | 67 | |
| CyclicReflexBackbone=Qwen3-8B, Decoding Strategy=CyclicReflex2025.06 | 63 | |
| OriginalBackbone=Qwen3-14B, Decoding Strategy=Original2025.06 | 63 | |
| TIPBackbone=Qwen3-14B, Decoding Strategy=TIP2025.06 | 63 | |
| Qwen3-30B-A3B-Thinking-2507Use Tool=false2026.05 | 61.5 | |
| CyclicReflexBackbone=Qwen3-4B, Decoding Strategy=CyclicReflex2025.06 | 61 | |
| TIPBackbone=Qwen3-8B, Decoding Strategy=TIP2025.06 | 60 | |
| TIPBackbone=Qwen3-4B, Decoding Strategy=TIP2025.06 | 59 | |
| OriginalBackbone=Qwen3-8B, Decoding Strategy=Original2025.06 | 59 | |
| OriginalBackbone=Qwen3-4B, Decoding Strategy=Original2025.06 | 58 | |
| qwen3-8B2026.01 | 57.29 | |
| PASC-GRPObase_model=qwen3-8B-Base2026.01 | 56.12 | |
| TRICE-4BUse Tool=true2026.05 | 55.6 | |
| BLASSTBackbone=Qwen3-8B, Attention Method=BLASST, Sparsity=~50%, Phase=Decode2025.12 | 54.15 | |
| GRP-SFTbase_model=qwen3-8B-Base2026.01 | 53.93 | |
| Dense AttentionBackbone=Qwen3-8B, Attention Method=Dense, Phase=Decode2025.12 | 53.83 | |
| RocketKVBackbone=Qwen3-8B, Attention Method=RocketKV, Phase=Decode2025.12 | 53.1 | |
| QuestBackbone=Qwen3-8B, Attention Method=Quest, Phase=Decode2025.12 | 52.17 | |
| Qwen3-4B-Thinking-2507Use Tool=false2026.05 | 51.4 | |
| BF16Model=Qwen3-4B, W-Bits=BF162026.01 | 48.38 | |
| Reasoning-QATModel=Qwen3-4B, W-Bits=W3G1282026.01 | 38.06 | |
| qwen3-8B-Base2026.01 | 29.76 | |
| GPTQModel=Qwen3-4B, W-Bits=W3G1282026.01 | 20.77 | |
| SKPO (Ours)Base Model=Qwen2.5-Math-7B2026.04 | 19.6 | |
| AWQModel=Qwen3-4B, W-Bits=W3G1282026.01 | 19.03 | |
| PRIMEBase Model=Qwen2.5-Math-7B2026.04 | 18.9 | |
| CISPOBase Model=Qwen2.5-Math-7B2026.04 | 17.4 | |
| DAPOBase Model=Qwen2.5-Math-7B2026.04 | 17 | |
| SAPOBase Model=Qwen2.5-Math-7B2026.04 | 16.5 | |
| GSPOBase Model=Qwen2.5-Math-7B2026.04 | 16.2 | |
| BF16Model=R1-Qwen-1.5B, W-Bits=BF162026.01 | 16.04 | |
| SPOBase Model=Qwen2.5-Math-7B2026.04 | 15.7 | |
| Critique†Base Model=Qwen2.5-Math-7B2026.04 | 15.2 | |
| GRPOBase Model=Qwen2.5-Math-7B2026.04 | 14.6 | |
| SKPO (Ours)Base Model=Llama-3.2-3B-Instruct2026.04 | 14.1 | |
| BaseBase Model=Qwen2.5-Math-7B2026.04 | 13.8 | |
| GSPOBase Model=Llama-3.2-3B-Instruct2026.04 | 13.8 | |
| PRIMEBase Model=Llama-3.2-3B-Instruct2026.04 | 13.2 | |
| GRPOBase Model=Llama-3.2-3B-Instruct2026.04 | 13.1 | |
| BF16Model=Qwen3-0.6B, W-Bits=BF162026.01 | 12.94 | |
| SPOBase Model=Llama-3.2-3B-Instruct2026.04 | 12.9 | |
| DAPOBase Model=Llama-3.2-3B-Instruct2026.04 | 12.7 | |
| SAPOBase Model=Llama-3.2-3B-Instruct2026.04 | 12.6 | |
| BaseBase Model=Llama-3.2-3B-Instruct2026.04 | 12.4 | |
| CISPOBase Model=Llama-3.2-3B-Instruct2026.04 | 12.1 | |
| Reasoning-QATModel=R1-Qwen-1.5B, W-Bits=W3G1282026.01 | 10.45 | |
| GPTQModel=R1-Qwen-1.5B, W-Bits=W3G1282026.01 | 9.33 | |
| AWQModel=R1-Qwen-1.5B, W-Bits=W3G1282026.01 | 4.85 | |
| Reasoning-QATModel=Qwen3-0.6B, W-Bits=W3G1282026.01 | 1.87 | |
| RTNModel=Qwen3-0.6B, W-Bits=W3G1282026.01 | 0 | |
| GPTQModel=Qwen3-0.6B, W-Bits=W3G1282026.01 | 0 | |
| AWQModel=Qwen3-0.6B, W-Bits=W3G1282026.01 | 0 | |
| GPTQModel=Qwen3-0.6B, W-Bits=W2G1282026.01 | 0 | |
| AWQModel=Qwen3-0.6B, W-Bits=W2G1282026.01 | 0 | |
| Reasoning-QATModel=Qwen3-0.6B, W-Bits=W2G1282026.01 | 0 | |
| RTNModel=R1-Qwen-1.5B, W-Bits=W3G1282026.01 | 0 | |
| GPTQModel=R1-Qwen-1.5B, W-Bits=W2G1282026.01 | 0 | |
| AWQModel=R1-Qwen-1.5B, W-Bits=W2G1282026.01 | 0 | |
| Reasoning-QATModel=R1-Qwen-1.5B, W-Bits=W2G1282026.01 | 0 | |
| RTNModel=Qwen3-4B, W-Bits=W3G1282026.01 | 0 | |
| GPTQModel=Qwen3-4B, W-Bits=W2G1282026.01 | 0 | |
| AWQModel=Qwen3-4B, W-Bits=W2G1282026.01 | 0 | |
| Reasoning-QATModel=Qwen3-4B, W-Bits=W2G1282026.01 | 0 |