Llama 3.3 70B
MetaTEXT Open weight
- Provider
- Meta
- Modality
- TEXT
- Parameters
- 70B
- Context window
- 128,000 tokens
- Weights
- Open
- Released
- 6 Dec 2024
Llama 3.3 70B Instruct is Meta's still-popular open-weight dense text model — GPT-4-class quality, 128K context, and the easiest large Llama to self-host.
Best for
- Proven, stable text workhorse for self-hosting
- Strong instruction-following and multilingual tasks
- Drop-in 70B dense model with broad tooling/quantization
- GPT-4-class text without Llama 4's deployment complexity
How it compares — and the India angle
- Text-only (no native vision) — choose Llama 4 Scout/Maverick for multimodal; choose 3.3 70B for a battle-tested text model.
- As a dense 70B it has the widest ecosystem and quantized builds — the easiest large Llama to run on modest Indian GPU setups.
- Open-weight and free to self-host keeps recurring costs near zero vs paid closed APIs — great for thin-margin Indian SaaS.
Benchmarks
Representative public scores (approximate, higher is better) for relative comparison. Check the provider for the latest official results.
MMLU86
GPQA Diamond51
How to access
Free to self-host (Llama 3.3 community licence). Dense 70B. Hosted cheaply (~$0.10–0.60 / 1M).
Access Llama 3.3 70BFrequently asked questions
What is Llama 3.3 70B?
Llama 3.3 70B is a text AI model developed by Meta with 70 billion parameters.
Who made Llama 3.3 70B?
Llama 3.3 70B was created by Meta.
Is Llama 3.3 70B open source?
Yes, Llama 3.3 70B has open weights and can be downloaded and self-hosted.
What is Llama 3.3 70B's context window?
Llama 3.3 70B supports a context window of 128,000 tokens.