On April 5, Meta announced the launch of its most advanced AI models with LIama 4 launch which aims to take a huge jump in the race for natively multimodal intelligence.
According to the official announcement, Meta introduced the first wave of its LIama 4 models, which include Scout, Maverick, and the experimental Behemoth. These new models are expected to deliver more personalized, cost-effective, and high-performing multimodal experiences.
With this launch, Meta will put itself forward in the race for AI models to compete with offerings from OpenAI, Google DeepMind, and Mistral.
LIama 4 Scout contains a 17-billion parameter model with 16 experts that fits entirely on a single NVIDIA H100 GPU. Scout provides a record-breaking 10 million token context window. According to the Meta team, it can outperform its competitors such as Gemma 3, Gemini 2.0 Flash-Lite, and Mistral 3.1 in benchmark tests.
The blog post states, “We’re making Llama 4 Scout and Llama 4 Maverick available for download today on llama.com and Hugging Face so everyone can continue to build new experiences using our latest technology. We’ll also make them available via our partners in the coming days. You can also try Meta AI with Llama 4 starting today in WhatsApp, Messenger, Instagram Direct, and on the Meta.AI website.”
Complementing Scout is LIama 4 Maverick, a more complex 17-billion active parameter model powered by 128 experts. Maverick has left observers in wonder by surpassing GPT-4o and Gemini 2.0 Flash on a wide range of tasks. It includes coding and reasoning, despite using fewer parameters.
It also scored an impressive 1417 ELO on LMArena, which is the highest score in the competitive performance of chat applications. Both models were distilled from Meta’s experimental LIama 4 Behemoth, which is a massive 288-billion parameter system still under training. Early results suggest Behemoth outpaces GPT-4.5 and Claude Sonnet 3.7 on STEM tasks.
Source: https://ai.meta.com/blog/llama-4-multimodal-intelligence/
Latest Stories:
