← Back to Directory
LLM2024-12-06
Llama 3.3 70B
Llama 3.3 70B provides Llama 3.1 405B level performance in a smaller package.
Overview
Llama 3.3 70B is a distilled version of Meta's largest models, providing frontier-level performance at a fraction of the size. It is the gold standard for 70B models.
Unique Factor
Matches Llama 3.1 405B performance in a model 6x smaller.
Key Capabilities
●
Distilled intelligence
●
128K context
●
Open weight
Benchmarks
MMLU Score
88%
HumanEval (Coding)
88%
GPQA Diamond
74%
MATH Benchmark
82%
Top Use Cases
Self-hosted Enterprise AI
Running a high-quality model on internal servers.
Example: “Generate a summary of this project proposal.”
Technical Specs
Context128,000 tokens
Params70B
LicenseLlama 3.3 Community
ArchTransformer
Developer
Meta's AI research division — makers of Llama, the world's most popular open-source LLM family.
Prompt Library
Browse Coding Prompts →
Previous Version
Llama 3 1 70b →