SmolLM3: smol, multilingual, long-context reasoner

Benchmark Model Rank Results
zero-shot-question-answering-on-openbookqaSmolLM3 3B Base#16Accuracy: 40.6
question-answering-on-boolqSmolLM3 3B Base#47Accuracy: 78.99
question-answering-on-piqaSmolLM3 3B Base#64Accuracy: 78.89
sentence-completion-on-hellaswagSmolLM3 3B Base#88Accuracy: 76.15
math-word-problem-solving-on-mathSmolLM3 3B Base (4-shot)#99Accuracy: 46.1
mathematical-reasoning-on-aime-2025SmolLM3 3B (Extended Thinking)#104Accuracy: 36.7
multiple-choice-question-answering-mcqa-on-gpqa-diamondSmolLM3 3B (Extended Thinking)#123Accuracy: 41.7
multi-task-language-understanding-on-mmlu-proSmolLM3 3B Base (MCF)#126Accuracy: 32.7
common-sense-reasoning-on-winograndeSmolLM3 3B Base#128Accuracy: 58.88
arithmetic-reasoning-on-gsm8kSmolLM3 3B Base (5-shot)#197Accuracy: 67.63