Meta just dropped Llama 4—on a weekend, can we ever catch a break? 😀—and there are two open weights models, Scout and M

Meta just dropped Llama 4—on a weekend, can we ever catch a break? 😀—and there are two open weights models, Scout and Maverick.

Scout is particularly impressive with its 10M token context (!!!), first in the industry. Quite an engineering accomplishment from Meta.

The models are already available on Fireworks AI and we started to put them through their paces—getting some impressive, Gemini-2.5-territory, performance.

Next week, we’ll set them up in our own lab—we recently added a new machine with 150GB of NVIDIA cards so we can’t wait to see how it performs on them.

The models are a big deal for self-hosted, you’re-in-control models that are customers keep on asking.

So expect a write up to understand the cost, security, and fine-tuning implications these models have for your business!

Anand Sahay Rob Dielemans Mayank Verma Preetpal Singh Klaudia Wachnio (Zdunczyk)

https://ai.meta.com/blog/llama-4-multimodal-intelligence/