Through an API

Open and closed models answer the same suite-standard 1.0 through their APIs, ranked together. 1 ranked model.

#ModelPass rate
1DeepSeek V4.1 Flash 1017Open weights100%

Model leaderboard

86 models in hard prompts ranked by LMArena score · updated Nov 11, 2026

#ModelScore
86Llama-3.2-1B-InstructMeta Llama1044
85granite-3.0-2b-instructIBM Granite1073
84granite-3.0-8b-instructIBM Granite1092
83Llama-3.2-3B-InstructMeta Llama1095
82granite-3.1-2b-instructIBM Granite1138
81granite-3.1-8b-instructIBM Granite1145
80Mixtral-8x22B-Instruct-v0.1Mistral AI1150
79QwQ-32B-PreviewQwen1162
78Llama-3.1-8B-InstructMeta Llama1175
77granite-4.0-h-smallIBM Granite1241
76Llama-3.1-70B-InstructMeta Llama1241
75Qwen2.5-Coder-32B-InstructQwen1251
74Llama-3.3-70B-InstructMeta Llama1257
73Llama-4-Scout-17B-16E-InstructMeta Llama1266
72Llama-3.1-405B-InstructMeta Llama1269
71Qwen2.5-72B-InstructQwen1271
70gpt-oss-20bOpenAI1274
69Llama-4-Maverick-17B-128E-InstructMeta Llama1281
68granite-4.1-8bIBM Granite1294
67granite-4.2-3bIBM Granite1304
Show all 86 models

Score by size

Plot against
Best for its size

Scores: LMArena, CC BY 4.0. Measured on served versions, not local builds. A model with several served variants shows its best.