Meta releases Llama 4: open-weight mixture-of-experts with 2T parameters
The flagship Llama 4 Behemoth model is the largest open-weight MoE ever released, with smaller Maverick and Scout siblings tuned for consumer hardware.
The flagship Llama 4 Behemoth model is the largest open-weight MoE ever released, with smaller Maverick and Scout siblings tuned for consumer hardware.
Mistral’s newest flagship is trained entirely in EU data centers and ships with a commercial license that strict regulated industries can actually accept.
A minimalist agent framework from Hugging Face is quietly becoming the default for teams who got tired of LangChain abstractions.
The runaway favorite local-LLM runtime gets first-class multi-model support, hot-swapping and a built-in OpenAI-compatible router.