available
Weights
Base and instruct/reasoning weights are public.
Inspect this claim’s source ↗Hugging Face / FAMILY RECORD
SmolLM3: license, size and what is open.
A compact multilingual reasoning model whose engineering blueprint, mixtures, training frameworks, and alignment process are published.
Sizes and download
3B
- SmolLM3-3B
- Total: 3B. Active: 3B. Source ↗
1.92 GB listed GGUF file · File listing ↗
Selected artifact: SmolLM3-3B / SmolLM3-Q4_K_M.gguf. Format: GGUF Q4_K_M. File evidence checked 2026-10-11.
Selected ggml-org conversion of Hugging Face SmolLM3, not a maker-published artifact. This 1.92 GB listing is rounded, not total runtime memory or a census of all quantizations.
Can my computer run it?
Compatibility, runtime memory and speed are untested by Bright.
Choose your computer and inspect documented candidates →Nine things to check
Source check: . Downloads, when known, carry their own check date.
Checked release: SmolLM3-3B
Sources checked 2026-09-19
- SmolLM3 release
- · Primary source
The profile describes this checkpoint, not every release carrying the family name. Developer documentation is not independent reproduction.
THE MODEL / THE TERMS
Know what you are picking up.
- Parameters
- 3B
- Architecture
- Dense decoder transformer with grouped-query attention and alternating NoPE layers.
- Modalities
- text
- Context
- 128K tokens
- Weights license
- Apache-2.0
- Commercial use
- Permitted by Apache-2.0; constituent training datasets have their own terms.
- Memory
- The base parameter size makes consumer deployment practical when quantized; no single official minimum is asserted.
- Known quantizations
- Community GGUF, MLX, ONNX, and bitsandbytes variants exist
Specifications are developer-reported. Primary release documentation ↗ · All sources and terms
NINE DIMENSIONS / NO OPENNESS SCORE
What is actually available?
Each component has its own evidence. “Not established” means this source review did not verify availability; it is not a claim that nothing exists.
available
Architecture
Architecture and ablations are described in detail.
Inspect this claim’s source ↗available
Inference code
Transformers-compatible inference and prompting instructions are public.
Inspect this claim’s source ↗available
Training code
Nanotron, Datatrove, and Lighteval code and configurations are linked.
Inspect this claim’s source ↗available
Training recipe
The three-stage data mixture and training blueprint are published.
Inspect this claim’s source ↗available
Data information
Sources and mixture percentages are disclosed by training stage.
Inspect this claim’s source ↗partial
Training data
The named public datasets are accessible, but the release is not a single immutable corpus bundle.
Inspect this claim’s source ↗available
Evaluation
Evaluation details and Lighteval tooling are published.
Inspect this claim’s source ↗available
Commercial use
The model is Apache-2.0 licensed.
Inspect this claim’s source ↗WHY THIS RELEASE MATTERS
What to explore.
- Small footprint
- Transparent recipe
- Long context
- Dual reasoning mode
WHAT THE RECORD DOES NOT ESTABLISH
Keep the limits.
- Text-only
- Six explicitly supported languages
- Smaller capacity than frontier-scale models
No independent capability study is attached; the release’s reproducibility artifacts make such study practical.
FOLLOW THE RECORD
Sources, not superlatives.
- SmolLM3: smol, multilingual, long-context reasoner ↗ (institution)
- SmolLM3-3B model card ↗ (registry)
Source check: 2026-09-19. AI-assisted editorial review, with no named external expert or independent hardware validation claimed. Licenses and model cards can change. Review the linked terms for the specific artifact and intended use.
Understand open source vs open weights ↗