Meta Llama 4

Llama 4 Maverick 17B 128E Instruct is an instruction-tuned, natively multimodal mixture-of-experts (MoE) large language model optimized for advanced instruction following and vision-language reasoning. It activates 17B parameters (400B total) across 128 experts and is built on a 48-layer transformer architecture with 40 attention heads and a 5,120 hidden size. The model supports multilingual text and image inputs and generates multilingual text and code outputs. With a 1M token context window, it is designed for large-scale multimodal workloads, enterprise AI systems, and long-context assistant-style applications.

Llama 4 Maverick 17B 128E is a natively multimodal mixture-of-experts (MoE) large language model engineered for high-performance text and image understanding. It activates 17B parameters (400B total) across 128 experts and is built on a 48-layer transformer architecture with 40 attention heads and a 5,120 hidden size. The model supports multilingual text and image inputs and produces multilingual text and code outputs. With up to a 1M token context window, it is optimized for scalable multimodal AI systems, large-document reasoning, and research-grade long-context workflows.

Llama 4 Scout 17B 16E is a natively multimodal mixture-of-experts (MoE) large language model built for advanced text and image understanding. It activates 17B parameters (109B total) across 16 experts and is built on a 48-layer transformer architecture with 40 attention heads and a 5,120 hidden size. The model supports multilingual text and image inputs and generates multilingual text and code outputs. With a 10M token context window, it is optimized for long-context reasoning, large-scale document analysis, and production-ready multimodal AI deployments.

Llama 4 Scout 17B 16E Instruct is an instruction-tuned, natively multimodal auto-regressive model developed by Meta, built on a mixture-of-experts architecture with 17B activated parameters (109B total). Designed for assistant-style and vision-language workloads, it supports multilingual text and image inputs and produces multilingual text and code outputs. The model features a native 10 million token context window, enabling extremely long-context analysis across documents, conversations, and multimodal inputs. It was trained on a mixture of publicly available and licensed data, including data from Meta’s products, with a knowledge cutoff of August 2024, and officially released on April 5, 2025.