Llama 3.3 70B
MetaTEXT Open weight
- Provider
- Meta
- Modality
- TEXT
- Parameters
- 70B
- Context window
- 128,000 tokens
- Weights
- Open
- Released
- —
Meta's open-weight 70B model offers strong general performance that you can download and self-host for free.
It is great for private, on-prem assistants and fine-tuning, and runs on a capable GPU or rented cloud GPU.
Best for
- Private, self-hosted assistants
- Fine-tuning on your own data
- Cost control at high, steady volume
- Offline and on-prem apps
Strengths & how it compares
- Best-known open-weight model with a huge community and tooling.
- Near closed-model quality for everyday tasks, at ₹0 licence cost.
- Vs DeepSeek V3 / Qwen 2.5: similar tier; pick by hardware and language needs.
Benchmarks
Representative public scores (approximate, higher is better) for relative comparison. Check the provider for the latest official results.
MMLU (knowledge)86
GPQA (reasoning)50
HumanEval (coding)80
How to access
Free (open weights); you pay only for hardware/GPU.
Access Llama 3.3 70B