Official benchmarks map / gsm8k
XiaomiMiMo/MiMo-V2.5-Pro is ranked first on the gsm8k leaderboard with 99.6, 2.9% ahead of meta-llama/Llama-3.1-405B. Read live from the Hugging Face API on November 8, 2026.
| # | Model | Organization | Score |
|---|---|---|---|
| 1 | XiaomiMiMo/MiMo-V2.5-Pro | XiaomiMiMo | 99.6 |
| 2 | meta-llama/Llama-3.1-405B | meta-llama | 96.8 |
| 3 | inclusionAI/LLaDA2.0-flash | inclusionAI | 96.06 |
| 4 | ibm-granite/granite-4.1-30b | ibm-granite | 94.16 |
| 5 | Motif-Technologies/Motif-3-Base | Motif-Technologies | 93.93 |
| 6 | meta-llama/Llama-3.2-90B-Vision-Instruct | meta-llama | 93.1 |
| 7 | deepseek-ai/DeepSeek-V4-Pro | deepseek-ai | 92.6 |
| 8 | ibm-granite/granite-4.1-8b | ibm-granite | 92.49 |
| 9 | SJTU-DENG-Lab/MBD-Math-LLaDA2-mini-16B | SJTU-DENG-Lab | 91.96 |
| 10 | microsoft/Phi-3-medium-4k-instruct | microsoft | 91 |
gsm8k is a dataset that Hugging Face marks as an official benchmark. Its leaderboard is built automatically from the .eval_results files that model repositories publish, so entries are mostly self-reported by model authors and evaluation settings differ between submissions. A verified tag appears only where Hugging Face marks an entry as verified.
Original board: huggingface.co/datasets/openai/gsm8k
Part of the Hugging Face official benchmarks map by QuantID. Unofficial. Not affiliated with or endorsed by Hugging Face.