Llama 3.3 (70B Instruct)
Llama · ~70B parameters · 13B+ · Last reviewed 2026-03-20
Why we use it
High-quality teacher and benchmark anchor for enterprises that need frontier-like behavior inside a private boundary before compressing to SLMs.
License summary
Meta Llama license; large checkpoints have higher infra and compliance overhead - ensure your deployment matches license redistribution rules.
Typical deployment profiles
- VPC, multi-GPU
- Datacenter / large clusters
Focus tags
- General
- Reasoning
- Multilingual
Typical use cases
- Teacher for distillation
- Private “full-size” assistant
- Difficult eval sets