Math on GSM-Plus
90.7ScoreQwen3.5-9B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3.5-9BParameters=9B, Variant=Thinking2026.05 | 90.7 | — | |
| LLaDA2.0-flash2026.02 | 89.74 | 268 | |
| Ling-flash-2.02026.02 | 89.71 | 100 | |
| LLaDA2.1-flashInference Mode=Q Mode2026.02 | 89.69 | 383 | |
| Qwen3-30B-A3B-Inst-25072026.02 | 89.41 | 100 | |
| Qwen3.5-4BParameters=4B, Variant=Thinking2026.05 | 89.3 | — | |
| LLaDA2.1-flashInference Mode=S Mode2026.02 | 89.23 | 714 | |
| OLMo-3-7BParameters=7B, Variant=Thinking2026.05 | 88.1 | — | |
| Ling-mini-2.02026.02 | 87.18 | — | |
| Mellum 2 (RL)Post-training Stage=RL, Parameters=2.5B/12B, Variant=Thinking2026.05 | 87 | — | |
| LLaDA2.1-minimode=Q Mode2026.02 | 86.55 | 3.69 | |
| LLaDA2.0-mini2026.02 | 86.5 | 2.41 | |
| Ministral-3-14BParameters=14B, Variant=Thinking2026.05 | 86.5 | — | |
| LLaDA2.1-minimode=S Mode2026.02 | 85.88 | 6.82 | |
| Qwen3-8Bno think=true2026.02 | 85.56 | — | |
| Full-FTModel=Qwen3-8B-Base, # Bits=16/16, Decode Execution=Independent2026.03 | 65.7 | — | |
| QSUNModel=Qwen3-8B-Base, # Bits=16/4, Decode Execution=Shared2026.03 | 64.9 | — | |
| AWQModel=Qwen3-8B-Base, # Bits=4/4, Decode Execution=Independent2026.03 | 64 | — | |
| Mellum 2 (SFT)Post-training Stage=SFT, Parameters=2.5B/12B, Variant=Thinking2026.05 | 62.6 | — | |
| Full-FTModel=LLaMA3.1-8B, # Bits=16/16, Decode Execution=Independent2026.03 | 49.8 | — | |
| Math Model# Model=1, KV Sharing=.2026.02 | 48.5 | — | |
| Multi Model# Model=3, KV Sharing=X2026.02 | 48.5 | — | |
| QSUNModel=LLaMA3.1-8B, # Bits=16/4, Decode Execution=Shared2026.03 | 48.1 | — | |
| AWQModel=LLaMA3.1-8B, # Bits=4/4, Decode Execution=Independent2026.03 | 47.6 | — | |
| ICaRus# Model=3, KV Sharing=O2026.02 | 45.8 | — | |
| Base Model# Model=1, KV Sharing=.2026.02 | 18 | — | |
| Coding Model# Model=1, KV Sharing=.2026.02 | 17.5 | — | |
| IF Model# Model=1, KV Sharing=.2026.02 | 16.5 | — |