In short
Llama 4 Maverick is Meta's open-weight Mixture-of-Experts model designed for complex scientific and mathematical reasoning. It activates only 64B parameters per inference to balance high performance with lower latency, making it suitable for private enterprise deployment.
Frequently asked
- What is the architectural difference between Llama 4 Maverick and previous Llama models?
- Maverick transitions from a dense transformer to a Mixture-of-Experts (MoE) framework. It has 400 billion total parameters but activates only approximately 64 billion during a single inference pass.
- How does the Reasoning Router function in Llama 4 Maverick?
- The Reasoning Router is a gating mechanism that identifies queries requiring multi-step logic. It directs these specific queries to expert layers optimized for symbolic logic and mathematical proofs.
- What are the licensing restrictions for using Llama 4 Maverick?
- The weights are released under the Llama Community License, allowing free download for research and commercial use. However, restrictions apply to organizations with over 700 million monthly active users.
- What is the context window size for Llama 4 Maverick?
- The model supports a massive 128k context window, which allows it to process entire codebases or books in a single pass.
- What are the API pricing rates for Llama 4 Maverick?
- Standard rates on major cloud providers are $0.75 per 1M input tokens and $1.50 per 1M output tokens. Custom pricing is available for Meta Managed Services or dedicated cloud instances.