Meta AI
Open-source frontier models democratising access to cutting-edge AI capabilities.
About
About Meta AI
Meta AI develops the Llama family of open-weight language models, providing competitive frontier-model performance under permissive licences. The Llama 4 generation introduced natively multimodal architecture and context windows reaching 10 million tokens in Scout. All Llama models are free to download, fine-tune, and deploy — eliminating API costs for teams with their own inference infrastructure.
Key highlights
Free open-weight models — no API costs, no usage restrictions for most use cases
Llama 4 Scout offers the largest context of any open model (10M tokens)
Natively multimodal MoE architecture in the Llama 4 generation
Supported by every major inference stack: Ollama, vLLM, llama.cpp, TGI
Available on Hugging Face, Meta AI website, and major cloud providers
Models
4 models tracked
Context windows, pricing, capabilities, and API identifiers — everything you need to choose the right model for your workload.
Llama 4 Maverick
2025
Meta's flagship open model. A 400B mixture-of-experts architecture with 17B active parameters per token. Competitive with GPT-4o on multimodal benchmarks — free to download and self-host.
- Open weights (Llama licence)
- Natively multimodal
- MoE architecture
- Competitive with GPT-4o
meta-llama/Llama-4-Maverick-17B-128E-InstructLlama 4 Scout
2025
The context champion of the Llama 4 family, supporting up to 10 million tokens — enough to fit entire codebases or document libraries in a single prompt.
- 10M token context
- Open weights
- Efficient MoE
- Entire codebase in context
meta-llama/Llama-4-Scout-17B-16E-InstructLlama 3.3 70B
2024
A proven workhorse with strong instruction following and coding capabilities. Supported by virtually every open-source inference platform: Ollama, vLLM, llama.cpp.
- Proven quality
- Universal hosting support
- Strong coding
- Fast inference
meta-llama/Meta-Llama-3.3-70B-InstructLlama 3.1 405B
2024
The largest publicly released dense language model. Used as a teacher model for distillation and as a quality benchmark for comparing smaller open models.
- Largest open dense model
- High quality ceiling
- Distillation source
- Multi-task benchmark leader
meta-llama/Meta-Llama-3.1-405B-InstructUse Cases
What Meta AI is best for
Each provider has specialisms shaped by their architecture decisions, training data, and API design. Here is where Meta AI performs best.
Get started
Start building with Meta AI
Visit the official Meta AI platform to access API keys, documentation, and pricing details.