Llama
last releaseApril 5, 2025
powered byLlama 4 Scout, Llama 4 Maverick
goblin vibe check:
if you're building something custom and don't want to be tied to an api bill forever, this is the obvious place to start
meta's open-weight model family for local, cloud, and embedded language and multimodal applications. llama 4 made the line natively multimodal with mixture-of-experts models and a huge context window, keeping it central for local ai builders even as newer proprietary meta work moved elsewhere.
context
10m
tokens
cost
free
Scout uses 17B active parameters with 16 experts and fits in a single H100-class deployment pathScout offers a 10M context window according to Meta's launch postMaverick uses 17B active parameters with 128 experts for stronger multimodal and reasoning workloadsmodels are available through Meta and partners including Hugging Face, Kaggle, edge partners, and cloud partners
key features
Scout uses 17B active parameters with 16 experts and fits in a single H100-class deployment pathScout offers a 10M context window according to Meta's launch postMaverick uses 17B active parameters with 128 experts for stronger multimodal and reasoning workloadsmodels are available through Meta and partners including Hugging Face, Kaggle, edge partners, and cloud partners
limitations
large variants still require serious serving hardware despite open availability
scope:
codelanguagemodelchatbotapilocalcloudbenchmark-strongvisualopen-source
launchFebruary 24, 2023
last releaseApril 5, 2025