Model
Model explorer

Llama 3.2 3B

OPEN
Meta · Llama 3.2 family · released Sep 25, 2024

Lightweight edge model for local summarization, rewriting and instruction following, with 128K context.

ReasoningCodingVisionFunction callingTool useAgentic
380.2
Elo · rank #336
Parameters
3.21B
Active params
3.21B (dense)
Context
128K tokens
Architecture
Dense decoder-only Transformer (GQA), distilled from Llama 3.1 8B/70B
License
Llama 3.2 Community License Agreement
Languages
8+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
ARC-ChallengeReasoning78.6%#40
best: Llama 3.1 405B · 96.9%
Arena EloHuman preference1166#45
best: Claude Fable 5 · 1505
BFCLAgents67.0%#24
best: SmolLM3 3B (No Thinking) · 92.3%
GPQA DiamondReasoning32.8%#251
best: GPT-5.6 · 94.6%
GSM8KMath77.7%#89
best: Llama 3.1 405B · 96.8%
HellaSwagReasoning69.8%#125
best: Claude 3 Opus · 95.4%
IFEvalReasoning77.4%#99
best: Gemma 4 26B A4B · 98.5%
MATH-500Math48.0%#139
best: GPT-5 · 99.4%
MGSMMath58.2%#42
best: OpenAI o4-mini · 93.7%
MMLUKnowledge63.4%#187
best: OpenAI o3 · 92.9%
TruthfulQAKnowledge51.5%#50
best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%
WinoGrandeReasoning71.1%#95
best: PaLM 2 · 90.9%
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
No widely reported RTX 4090 llama.cpp Q4 benchmark found; expected well over 100 tok/s given its small size
Quantizations
GGUF Q4 · AWQ · GPTQ · MLX
Llama 3.2 family
Elo progression across releases
Llama 3.2 11B VisionSep 2024682.7Llama 3.2 1BSep 202481.3Llama 3.2 3BSep 2024380.2Llama 3.2 90B VisionSep 2024999.5
Resources
HFmeta-llama/Llama-3.2-3B-InstructGHmeta-llama/llama-modelsDOCSllama.com/docs/model-cards-and-prompt-formats/llama3_2APIAvailable via Meta API
API price weights · each benchmark row carries its own source badge (see methodology)