GSM8K-Platinum
Math · 2025-03-06
GSM8K-Platinum is a cleaned revision of the GSM8K test set in which questions with ambiguous wording or wrong reference answers were relabeled or removed, leaving 1,209 grade-school math word problems graded by final-answer match. Scores report accuracy; shot count is recorded per row. It is a fallback component of the GSM8K group. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| Kimi K2 Base | 94.2 |
| DeepSeek-V3-Base | 93.4 |
| Qwen2.5 72B | 92.5 |
| Kimi Linear 48B A3B Base | 89.6 |
| Llama 4 Maverick | 88.8 |
| Moonlight 16B A3B | 79.4 |