Mathematical Reasoning on GSM8K (Original, Emotional, Neutralized Accuracy) benchmark leaderboard · SOTA2 Research

Mathematical Reasoning on GSM8K (Original, Emotional, Neutralized Accuracy)

98Original AccuracyGPT-4o

Evaluation Results

Method	Links
GPT-4o2026.04		98	95	95.3
GPT-4o2026.04		97.8	95.2	95.2
Llama-3.3-70B2026.04		97.5	93.2	94.7
GPT-52026.04		97.5	94.5	94.9
GPT-5.42026.04		97.5	96	95.7
o32026.04		97.5	94.6	95.2
GPT-5.42026.04		97	95.8	95.8
o32026.04		97	94.4	95.7
Llama-3.3-70B2026.04		96.8	93.7	94.6
Qwen-2.5-72B2026.04		96.8	93.9	94.3
GPT-52026.04		96.5	94.9	95.2
Frontier avg2026.04		96.5	94.2	95
Frontier avg2026.04		96.4	94	94.5
DeepSeek-V32026.04		95.8	94.1	94.5
DeepSeek-V32026.04		95.8	93.7	93.9
Mistral-Sm-24B2026.04		95.5	91.7	93.1
Qwen-2.5-72B2026.04		95.2	93.7	94.4
Mistral-Sm-24B2026.04		95	91.8	93.8
GPT-4o-mini2026.04		94.8	90.9	93.3
Qwen-2.5-7B2026.04		92.8	88	89.7
GPT-4o-mini2026.04		92.5	90.2	92.3
Gemma-2-27B2026.04		92	89.2	90.5
Gemma-2-27B2026.04		91.8	87.8	90.6
Gemma-2-9B2026.04		88.8	85.4	88.5
Gemma-2-9B2026.04		87	83.8	86.5
Qwen-2.5-3B2026.04		86.2	79.1	83.9
Qwen-2.5-3B2026.04		86	81.1	85
Llama-3.1-8B2026.04		86	80.9	84.4
Llama-3.1-8B2026.04		84	78.5	82.6
Open-src avg2026.04		80.9	76.7	79.6
Qwen-2.5-7B2026.04		80.2	73.7	78.8
Open-src avg2026.04		78.6	73.5	77.4
Llama-3.2-3B2026.04		76.2	67.6	74.2
Llama-3.2-3B2026.04		74.5	64.5	71.5
Gemma-2-2B2026.04		64	59.2	62.5
Mistral-7B2026.04		61.2	57	58.7
Gemma-2-2B2026.04		60.8	56.9	61.7
Mistral-7B2026.04		56	49.5	53.8
Llama-3.2-1B2026.04		35.2	29	36.8
Llama-3.2-1B2026.04		34	33.4	39.7