Llama 4 Maverick (and Llama 4 Scout)
Llama 4 Scout and Maverick, released April 5, 2025, are still the newest Llama models and remain widely used open-weight options. They are mixture-of-experts models that read text and images. Meta has since moved its main development to the Muse…
Best for
Existing self-hosted deployments that already run Llama and need a mixture-of-experts model with very long context.
Overview
Llama 4 Scout and Maverick, released April 5, 2025, are still the newest Llama models and remain widely used open-weight options. They are mixture-of-experts models that read text and images. Meta has since moved its main development to the Muse family.
Strengths and limits
- Mixture-of-experts: 17B active parameters (400B total for Maverick, 109B for Scout)
- Scout's 10M-token context window
- Huge ecosystem of tooling and fine-tunes
- Custom community license with conditions, not Apache 2.0
- Outpaced by 2026 models; Meta's new work is under the Muse brand
Published benchmark scores
Scores are as published by the source; settings (reasoning effort, tools, attempts) differ between labs, so compare with care. How to read benchmarks →