Model
Model explorer

SmolLM3 3B (Think)

OPEN
Hugging Face · SmolLM family · released Jul 8, 2025

Default extended-thinking mode (/think); open 3B multilingual dual-mode reasoner with 128K context, trained on 11.2T tokens.

ReasoningCodingVisionFunction callingTool useAgentic
881.6
Elo · rank #241
Parameters
3.08B
Active params
3.08B (dense)
Context
128K tokens
Architecture
Decoder-only Transformer with GQA + NoPE (selective RoPE removal) + intra-document masking
License
Apache-2.0
Languages
6+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
AIMEMath36.7%#129
best: GPT-5.2 · 100.0%
BFCLAgents88.8%#4
best: SmolLM3 3B (No Thinking) · 92.3%
GPQA DiamondReasoning41.7%#211
best: GPT-5.6 · 94.6%
IFEvalReasoning71.2%#112
best: Gemma 4 26B A4B · 98.5%
LiveCodeBenchCoding30.0%#147
best: DeepSeek-V4-Pro (Think Max) · 93.5%
Run it locally
VRAM @ Q4
2.5 GB
VRAM @ FP16
6.5 GB
Fits on (Q4)
RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Throughput data unavailable.
Quantizations
GGUF Q4 · MLX
SmolLM family
Elo progression across releases
SmolLM2 1.7BNov 2024-93.8SmolLM3 3B (No Thinking)Jul 2025703.6SmolLM3 3B (Think)Jul 2025881.6
Resources
HFHuggingFaceTB/SmolLM3-3BGHhuggingface/smollmDOCShuggingface.co/blog/smollm3
API price weights · each benchmark row carries its own source badge (see methodology)