SmolLM3-3B
Fully open 3B model with dual-mode reasoning and tool calling.
About
Hugging Face's 3B model built to show what careful engineering can squeeze out of a small footprint. It reasons in two modes — an extended thinking trace by default, or a /no_think flag for fast direct answers — and ships with XML and Python-style tool calling.
Everything is open: weights, the public data mixture, and the training configs, all under Apache 2.0. One of the most transparent small-model releases to date.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Outperforms Llama 3.2 3B and Qwen2.5 3B; competitive with 4B models (Qwen3, Gemma 3) per Hugging Face.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Hugging Face org release, widely downloaded and fine-tuned across the open ecosystem; Active maintenance from the SmolLM team; base + instruct + checkpoints all published; Weights, public data mixture, and training configs all released under Apache 2.0. Manual review: dual-mode reasoning and tool calling verified on the HF model card and repo README. License: Apache 2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Hugging Face org release, widely downloaded and fine-tuned across the open ecosystem
- · Active maintenance from the SmolLM team; base + instruct + checkpoints all published
- · Weights, public data mixture, and training configs all released under Apache 2.0
- · manual review: dual-mode reasoning and tool calling verified on the HF model card and repo README