Meta Unveils Llama 4: New Open-Source AI Models for Enhanced Multimodal Capabilities

Meta Releases Llama 4: A New Generation of Open-Source AI Models
Meta has announced the latest iteration of its open-source AI model family, Llama 4, amidst intensifying competition in the generative AI industry. This new family includes four models, with Meta detailing Llama 4 Scout, Llama 4 Maverick, and Llama 4 Behemoth. These models were trained on extensive unlabeled text, image, and video data, indicating a strong focus on multimodal capabilities.
Key Models and Their Capabilities:
- Llama 4 Scout: Designed for efficiency, capable of running on a single Nvidia H100 GPU.
- Llama 4 Maverick: Positioned for creative tasks like coding and reasoning, outperforming models like OpenAI's GPT-4o and Google's Gemini 2.0 in specific benchmarks. However, it faces stiff competition from advanced models like Gemini 2.5 Pro, GPT-4.5, and Anthropic's Claude 3.7 Sonnet.
- Llama 4 Behemoth: The most powerful model in the family, currently in training. Meta aims for it to significantly outperform comparable models and act as a teacher for other Llama 4 models. However, reducing the hardware costs for training this model remains a challenge.
Availability and Access:
Currently, Llama 4 Scout and Llama 4 Maverick are accessible to users through Meta's platforms, including WhatsApp, Messenger, and Instagram Direct, as well as Meta's AI website, Llama.com. Developers can also access these models via open-source repositories like Hugging Face.
Competitive Landscape and Strategy:
Meta's release of Llama 4 is a strategic move to compete with major players like OpenAI and Google. The company has been closely studying competitors, including Chinese AI company DeepSeek, known for its cost-effective models. Meta aims to leverage its open-source approach to foster innovation and adoption within the AI community.
Performance Benchmarks and Comparisons:
Meta has conducted internal testing to compare Llama 4 models against industry benchmarks. While Maverick shows promise in creative writing and coding, it lags behind some of the leading models in certain areas. The company is focused on improving the performance and cost-efficiency of its models, particularly the Behemoth variant.
Future Developments and Events:
Meta is set to host its first LlamaCon AI conference on April 29th, focusing on open-source AI. Additionally, a standalone Meta AI chatbot is slated for launch in the second quarter of the year.
Industry Trends and Competitor Moves:
Meta's strategy mirrors that of OpenAI, which has also adjusted its release timelines for models like GPT-5 to ensure quality and performance. The industry is witnessing a rapid evolution of AI capabilities, with a strong emphasis on multimodal understanding and efficient model deployment.
Author and Related Content:
This article was written by Fionna Agomuoh, a Computing Writer at Digital Trends. The content is supplemented with links to related articles on AI, computing, and entertainment, providing readers with a comprehensive overview of the AI landscape.
Image: Meta AI widget on Home Screen.
Image: Meta Llama 4 training information.
Related Articles:
- Meta made insane offers in bid to nab OpenAI talent, Altman claims
- OpenAI makes its most advanced coding model available to paid ChatGPT users
- Meta’s new AI app lets you share your favorite prompts with friends
- Ray-Ban Meta Glasses are my favorite AI gadget, and they keep getting better
- It’s not your imagination – ChatGPT models actually do hallucinate more now
- The original AI model behind ChatGPT will live on in your favorite apps
- Your politeness toward ChatGPT is increasing OpenAI’s energy costs
- Meta is training AI on your data. Users say opting out doesn’t work.
More Computing News:
- The new Reachy Mini robot can let kids turn play into innovation
- Apple loses AI whiz to Meta with an offer that will make your eyes water
- Watch how NASA is using VR to prep for upcoming moon missions
- This Dyson doesn’t suck … it grows
- Wimbledon’s robot line judges caught napping on Centre Court
Original article available at: https://www.digitaltrends.com/computing/metas-latest-open-source-ai-models-challenge-gpt-gemini-and-claude/