Llama 4 Scout

Llama 4 Scout, released by Meta in April 2025, is an open-weight model utilizing a Mixture of Experts (MoE) architecture with 109B total parameters but only 17B active parameters during inference. This design provides strong capabilities while maintaining computational efficiency. The model's standout feature is its industry-leading 10 million token context window-the largest context capacity of any production language model.

Llama 4 Scout beats competing models like Gemma 3, Gemini 2.0 Flash-Lite, and Mistral Small 3.1 across various benchmarks, with overall performance comparable to GPT-4o mini. This positions it as a strong open-source alternative to proprietary small models, offering similar capabilities without API costs or data privacy concerns.

The massive 10M token context window enables entirely new use cases, such as processing multiple full-length books, entire codebases with all dependencies, or maintaining context across very long conversations. Combined with its open-weight nature and efficient MoE architecture, Llama 4 Scout represents a significant contribution to the open AI ecosystem, particularly for applications requiring extreme context lengths.