SmolLM3 3B
HuggingFaceTB/SmolLM3-3B
Open Source · chat · open-weights
Context
128K
Max output
—
Pricing
Open weights
Modalities
text
Released
08 Jul 2025
License: apache-2.0 · HuggingFaceTB/SmolLM3-3B
AI summary
● machine-written
SmolLM3 3B: Open-source 3B model with 128K context and multilingual support
SmolLM3 is a 3 billion parameter decoder-only transformer from Hugging Face designed for efficient deployment in resource-constrained environments. Trained on 11.2 trillion tokens, it supports up to 128K token context length and native reasoning capabilities across 6 languages. The model achieves performance comparable to much larger models through optimized architecture and high-quality training data.
What's new
- 128K token context window with YaRN extrapolation support
- Trained on 11.2T tokens using three-stage curriculum mixing web, code, and math data
- Native multilingual support for 6 languages with consistent reasoning quality
- Dual-mode intelligence with /think and /no_think reasoning modes
- Apache 2.0 licensed open-weights model
Best for
Edge deployment on mobile devices and constrained environmentsLong document analysis and research paper processingMultilingual customer service and cross-lingual applicationsIntelligent agent workflows with tool integration