Meta Rolls Out 3 AI Models, Llama 4, Claimed to Beat ChatGPT and Gemini

TEMPO.CO, Jakarta - Meta officially launched their latest range of artificial intelligence models, Llama 4, which are now installed in the Meta AI assistants on WhatsApp, Messenger, Instagram, and the web version.

The AI model consists of Llama 4 Scout, Llama 4 Maverick, and Llama 4 Behemoth. According to The Verge, the first two models can be downloaded through Meta or Hugging Face.

Meta found this in several commonly reported metrics, Llama 4 Scout is a lightweight model that can be loaded into a single Nvidia H100 GPU. With a context window of 10 million tokens, Meta claims Scout outperforms Google Gemma 3, Gemini 2.0 Flash-Lite, and Mistral 3.1.

On the other hand, Llama 4 Maverick is developed as a larger and tougher model, on par with GPT4.0 and Gemini 2.0 Flash. Meta notes that Maverick also performs better than these two AI models. Maverick's capabilities are said to be equivalent to DeepSeekV3 in programming and reasoning tasks, while using less than half of the active parameters.

The third model, Llama 4 Behemoth, is still in training. This model has 288 billion active parameters out of 2 trillion. Meta's CEO Mark Zuckerberg refers to Behemoth as “the highest performing base model in the world.” Meta also claims Behemoth can outperform GPT4.5 and Claude Sonnet 3.7 in some STEM metrics.

All Llama 4 models are built using the mixture of experts (MoE) architecture, an efficient approach that only activates the parts of the model needed for specific tasks.

Although promoted as open-source, Llama 4 still has licensing limitations. Commercial entities with more than 700 million monthly active users must seek permission from Meta before using this model. According to the Open Source Initiative, this provision removes Llama 4 from the Open Source product category.

Meta will reportedly publicly discuss the development of AI in its products. One agenda item the company will use for this plan is the LlamaCon conference on April 29.

Editor's Choice: YouTube Launches New Editing Tools for Shorts to Rival TikTok and Reels

Click here to get the latest news updates from Tempo on Google News

Top AI Applications of 2024: Beyond ChatGPT and NovelAI

19 jam lalu

OpenAI Invests in Adaptive Cybersecurity

1 hari lalu

Elon Musk Takes Over X Through xAI, Valuing It at Rp546 Trillion

7 hari lalu

OpenAI's Ghibli-Style AI Images Spark Copyright Concerns and Artists' Outrage

9 hari lalu

Google Claims Gemini 2.5 as Its Most Intelligent AI Yet

10 hari lalu

OpenAI Unveils GPT-4o: AI for Image Creation and Editing

10 hari lalu

North Korea Leader Kim Jong Un Oversees Tests of AI Suicide Drones

10 hari lalu

Apple's Base iPad Won't Include AI; Here's Why

11 hari lalu

OpenAI Updates ChatGPT's Advanced Voice Mode: What's New and How It Works?

11 hari lalu

Apple Watch to Feature AI-Powered Camera, Launch Expected in 2027

11 hari lalu