← LLM Models#35Open SourceView on HF →
Llama 3.2 3B
Meta
Tiny but capable. Runs on-device (phone/laptop). 500+ tok/s on consumer GPU.
Official page →📖Context128K tokens
🚀Speed500 tok/s
💵Input price-
💸Output price-
🧠Parameters3B
Benchmarks
Gray bar = dataset averageMMLU57 academic subjects
63.4%avg 81.8
GPQA DiamondPhD-level science questions
24.7%avg 50.4
HumanEvalPython code generation
58.3%avg 79.9
SWE-Bench VerifiedReal GitHub engineering tasks
9.5%avg 32.4
Composite ScoreScore = MMLU×20% + GPQA×30% + HumanEval×25% + SWE-Bench×25%37.0
HF Open LLM Leaderboard
IFEval · BBH · MATH · GPQA · MuSR · MMLU-Pro
Average
24.2
IFEvalInstruction following
73.9%avg 79.9
BBHBig Bench Hard
24.1%avg 48.7
MATH Lvl 5Competition mathematics
17.7%avg 37.8
GPQAGraduate-level science Q&A
3.8%avg 12.9
MuSRMultistep soft reasoning
1.4%avg 10.7
MMLU-ProProfessional knowledge
24.4%avg 40.0
More from Meta
API Providers
Details
- Provider
- Meta
- Parameters
- 3B
- Context
- 128K tokens
- Speed
- 500 tok/s
- Open Source
- Yes
- License
- Open (varies)
- Released
- -