Meta Overhauls Llama 4 Series Amid Launch Delays and Updates

Meta is revamping its Llama series with notable changes including a shift to MoE architecture and native multimodality as earlier models (Scout and Maverick) are retired. Internal overhauls and developmental delays have led to the preview of Llama 4 Behemoth, setting the stage for a refreshed launch strategy.


11526 techinasia.com / Meta launches new AI models with advanced features

Supporting over 200 languages, these models are compatible with platforms like WhatsApp, Messenger, and Instagram Direct.

11448 theverge.com / Meta AI gets two new models as Meta releases Llama 4

Meta introduced Llama 4, its newest AI model suite powering Meta AI on the web, WhatsApp, Messenger, and Instagram Direct. Both models—including the compact Llama 4 Scout—are now available for download from Meta and Hugging Face.

11429 simonwillison.net / Initial impressions of Llama 4

Meta AI’s Llama 4 release features Scout and Maverick—Maverick is second on LM Arena behind Gemini 2.5 Pro. Try them via OpenRouter. Scout claims a 10M token window, though current limits are lower. System prompts reveal nuanced behavioral guidelines.

11449 venturebeat.com / Meta’s answer to DeepSeek is here: Llama 4 launches with long context Scout and Maverick models, and 2T parameter Behemoth on the way!

While DeepSeek R1 and OpenAI o1 edge out Behemoth on a couple metrics, Llama 4 Behemoth remains highly competitive.

11387 simonwillison.net / Note on 5th April 2025

Llama 4 models Maverick and Scout, released by Meta, are shaking up the leaderboard. Available via OpenRouter (chat/API), they feature varying token limits, with providers like Groq, Fireworks, and Together in play. These giant MoE models aren’t built for consumer GPUs but may run well on Macs.

11378 techcrunch.com / Meta releases Llama 4, a new crop of flagship AI models

Meta released Llama 4 AI models—including Scout, Maverick, and Behemoth—trained on vast unlabeled text, image, and video data to provide broad, multi-modal capabilities.

11355 simonwillison.net / Quoting Ahmed Al-Dahle

Meta's Llama series now uses MoE architecture with native multimodality. Llama 4 Scout and Maverick are retired, while Llama 4 Behemoth is previewed—delivering top STEM performance and outpacing leading LLMs.

11150 the-decoder.com / Meta reportedly prepares to launch Llama 4 after delays and internal overhaul

Meta’s next language model, Llama 4, is reportedly nearing release after repeated delays and significant changes to its development process. The article Meta reportedly prepares to launch Llama 4 after delays and internal overhaul appeared first on THE DECODER.


8 stories from sources in 21.3 hour(s) #ai #machine-learning #meta



Disclaimer: The information provided on this website is intended for general informational purposes only. While we strive for accuracy, we do not guarantee the completeness or reliability of the content. Users are encouraged to verify all details independently. We accept no liability for errors, omissions, or any decisions made based on this information.