Model
Model explorer

SmolLM3 3B (No Thinking)

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

Non-reasoning (/no_think) mode of SmolLM3; faster inference, still outperforms other 3B non-reasoning instruct models.

ReasoningCodingVisionFunction callingTool useAgentic
703.6
Elo · rank #266
Parameters
3.08B
Active params
3.08B (dense)
Context
128K tokens
Architecture
Dense transformer decoder, GQA (4 groups) + NoPE/RoPE 3:1 ratio, tied embeddings; single checkpoint with think/no_think dual reasoning modes
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
AIMEMath9.3%#148
best: GPT-5.2 · 100.0%
BFCLAgents92.3%#1
best: this model · 92.3%
GPQA DiamondReasoning35.7%#240
best: GPT-5.6 · 94.6%
IFEvalReasoning76.7%#102
best: Gemma 4 26B A4B · 98.5%
LiveCodeBenchCoding15.2%#166
best: DeepSeek-V4-Pro (Think Max) · 93.5%
Run it locally
VRAM @ Q4
3 GB
VRAM @ FP16
6 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
~300 tok/s on RTX 4090 (Q4, llama.cpp)
Quantizations
GGUF Q4 · MLX · ONNX
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/smollm
API price weights · each benchmark row carries its own source badge (see methodology)