Open LLM Leaderboard v1 - GSM8K: leaderboard

Original Hugging Face Open LLM Leaderboard grade-school arithmetic word problems, with five demonstrations and the harness answer-accuracy metric. Percentage, higher is better. Frozen v1 archive; genuine zero scores are retained and public questions may overlap model training.

Metric: Accuracy (%) (5-shot). Source: huggingface.co. Saturation forecast: Estimated already saturated. 7252 models tracked.

Top models

#ModelScoreOverall rank
1Llama 3 70B Japanese Suzume Vector V0.187.41#774
2Llama3-TenyxChat-70B86.28#738
3autotrain-llama3-70B-orpo-v185.67#842
4Llama 3 70B Instruct85.44#641
5Qwen 2 72B85.29#516
6Llama 3 70B Fireplace83.62#846
7Mixtral 8x22B Instruct82.03#740
8Qwen 1.5 110B81.05#734
9Le_Triomphant-ECE-TW379.83#563
10Smaug-72B-v0.178.7#613
11Llama 3 70B76.88#634
12Rhea-72B-v0.576.12#553
13free-evo-qwen72B-v0.8-re75.89#442
14YiSM-34B-0rn75.82#713
15SeaLLM-7B-v2.574.37#1065

No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #6 GPT-6.1 Sol.

Interactive version: theaggregate.ai/benchmark?slug=open-llm-leaderboard-v1-gsm8k · How It Works · Data refreshed daily, snapshot 2026-11-06.