Llama 4 Maverick Review

Meta's last major Llama-branded release, published on 5 April 2025 as the mid-size half of a pair alongside Scout. It is a mixture-of-experts model with 17 billion active parameters against 400 billion total, natively multimodal for text and image input with text output, trained across twelve languages, addressing a million tokens of context where Scout reached ten million. Weights are source-available rather than open source: the Llama 4 Community License carries acceptable-use terms and a scale threshold, a distinction that mattered a great deal more once genuinely Apache-licensed competitors of similar size arrived.

Its historical importance is hard to overstate, since the Llama line seeded most of the open fine-tuning ecosystem and remains the base for a great many derivative models. Meta has since moved on, replacing the Llama line with Muse Spark from Meta Superintelligence Labs in April 2026. This grade is a provisional placeholder pending hands-on testing.