LastChatMeta recently released two new models, Llama 4 Scout and Llama 4 Maverick which are now available in LastChat. Scout is a 109B parameter model, and Maverick is ~400B. Both models use a mixture-of-experts architecture which makes their answers fast and efficient.
One of the headlining features of these models is their long context support. Maverick has a context length of 1M tokens, while Scout has a shocking 10M token context length — an industry first. Unfortunately, you won't be able to try this out today as major model providers don't currently support this context length.
In addition, some benchmarks show that Llama 4 Scout and Maverick become less effective with longer texts, with their ability to comprehend and recall information dropping by about half when text exceeds 4,000 tokens. This is equivalent to ~3,000 words, or a medium-length news article.
I've been testing out both Scout and Maverick and they seem to have solid conversational skills, but at least in my experience are less capable in terms of instruction following.
Try Llama 4 in LastChat today and let us know what you think!
To learn more about the models, check out Meta's blog post.