👁 HuggingFace
HuggingFace
SmolLM3 3B
649.0KDownloads981LikesJul 2025Released128K tokensContextApache 2.0License21 EntryQuality
SmolLM3 3B (3B parameters) requires approximately 5.6 GB of VRAM with Q4_K_M quantization. For the best balance of quality and speed, we recommend hardware with at least 7 GB of VRAM.
Get started
— copy & paste to run locallyCopy-paste commands to run SmolLM3 3B on your machine.
Run
lms load SmolLM3-3B && lms server startQuick specs
Parameters3B
Architecturedense
Context128K tokens
Modalitytext
Min RAM1.2 GB
Rec. RAM1.8 GB (Q4_K_M)
LicenseApache 2.0
FamilySmolLM
✓ Chat✓ Reasoning
About this model
- •Dual-mode reasoning: extended thinking can be toggled on/off
- •128K context via YARN extrapolation from 64K training
- •6 natively supported languages: English, French, Spanish, German, Italian, Portuguese
- •Fully open: weights, training details, and public data mixture
Your hardware
Detecting...
Quick picks
Best hardware
Top picks for SmolLM3 3B
Run this model
Quantization options
VRAM estimates by quant level
No hardware detected — fit column shows raw VRAM estimates
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 1.2 GB | Low | — |
Q3_K_S | 3 | 1.5 GB | Low | — |
NVFP4 | 4 | 1.7 GB | Medium | — |
Q4_K_M | 4 | 1.8 GB | Medium | — |
Q5_K_M | 5 | 2.2 GB | High | — |
Q6_K | 6 | 2.5 GB | High | — |
Q8_0 | 8 | 3.2 GB | Very High | — |
F16 | 16 | 6.1 GB | Maximum | — |
Quality benchmarks
SmolLM3 3B benchmark scores
Coding
SWE-bench Verified—
HumanEval+30.5%
Aider Polyglot—
LiveCodeBench—
Reasoning
MMLU-Pro32.7%
GPQA Diamond35.7%
MATH-500—
ARC Challenge—
General
Chatbot Arena—
IFEval76.7%
Source: official · 2025-07-02
Hardware compatibility
Fit estimates across all hardware
Computing compatibility...
Memory breakdown
Reference: RTX 2060 6GB
Weights1.8 GB
KV Cache2.0 GB
Runtime1.2 GB
Headroom0.6 GB
Frequently asked questions
FAQ — SmolLM3 3B
See also