Meta Platforms has released its first Llama 4 models, a multimodal trio that ranges from the foundational Behemoth to tiny Scout, with Maverick in between. With 16 experts and only 17B active parameters (the number used per task), Llama Scout is “more powerful than all previous generation Llama models, while fitting in a single Nvidia H100 GPU,” according to Meta. Maverick, with 17B active parameters and 128 experts, is touted as beating GPT-4o and Gemini 2.0 Flash across various benchmarks, “while achieving comparable results to the new DeepSeek v3 on reasoning and coding with less than half the active parameters.”

When a Chinese firm released the open source DeepSeek reasoning model in January, “the entire AI landscape shifted,”  VentureBeat says, contending “Meta was reportedly sent into panic mode upon learning that this new R1 model had been trained for a fraction of the cost of many other leading models, as little as several million dollars — what it pays some of its own AI team leaders — yet still achieved top performance in the open source category.”

The move “forced some kind of reckoning” at Meta which, having released Llama 3.3 a month prior, found it “already looking outdated,” per VentureBeat, which calls the Llama 4 models “the fruits of that reckoning.”

With 2-trillion total parameters — the number used in training, amounting to the total sum of its knowledge — Llama 4 Behemoth is massive. It was used to train the other two, even though Meta explains in a blog post that it is still being trained itself and is only in preview while its distilled juniors are available on Llama.com and on Hugging Face, as well as through third-parties including Microsoft Azure and Amazon Bedrock.

“Meta says that Meta AI, its AI-powered assistant across apps, including WhatsApp, Messenger, and Instagram, has been updated to use Llama 4 in 40 countries,” reports TechCrunch, noting “multimodal features are limited to the U.S. in English for now.”

“Llama 4 will help power AI agents, which will be capable of new levels of reasoning and action, Meta Chief Product Officer Chris Cox said in March,” writes CNBC.

Related:
Meta’s Llama 4 Models Now Available on Amazon Web Services, AWS, 4/5/25
Meta’s Llama 4 Is Now Available on Workers AI, Cloudflare, 4/6/25

UPDATE:
Meta Got Caught Gaming AI Benchmarks, The Verge, 4/7/25