Hugging Face
Hugging Face's SmolLM2 1.7B instruction model — punches above its weight for on-device use, outperforming larger models on reasoning tasks.
Hugging Face's SmolLM3 3B model — compact, Apache 2.0 licensed, trained for strong reasoning and instruction-following at small scale.
SmolLM3 proxy — GPT4-Turbo baseline reference
