Today, Meta AI announced the release of its latest generation of multimodal models, Llama 4, with two variants: Llama 4 Scout and Llama 4 Maverick. These models represent significant technical progress in multimodal AI, providing improved opportunities for both text and image understanding.
The Llama 4 Scout is a 17-billion-active parameter model structured with 16 expert modules. It introduces an extensive context window capable of accommodating up to 10 million tokens. This significant context capacity allows the model to control and interpret extensive text content effectively, advantageous for long -shaped document processing, complex code bases and detailed dialogue tasks. In comparative evaluations, the Llama 4 scout has shown superior performance compared to modern models such as Gemma 3, Gemini 2.0 Flash-Lite and Mistral 3.1 across recognized benchmark data sets.
Parallel to Scout, Llama 4 Maverick, which is also based on a 17-billion-active parameter architecture, contains 128 expert modules that are explicitly designed to improve visual grounding. This design facilitates precise adjustment between textual PROMPS and affiliated visual elements, enabling targeted answers that are based exactly to specific image regions. Maverick exhibits robust performance in comparative assessments and surpasses GPT-4O and Gemini 2.0 flash, especially in multimodal reasoning tasks. In addition, Maverick has achieved comparable results with Deepseek V3 on reasoning and coding of benchmarks, while using about half of the active parameters.
An important feature of Maverick is its remarkable efficiency for costs. Benchmarking efforts, specifically on the LMARENA platform, have recorded an ELO assessment of 1417 for Maverick’s chat optimized version, indicating its calculation efficiency and practical usefulness in conversation and multimodal contexts.
The development of scouting and maverick is strongly drawing from distillation techniques that stem from the ongoing training of Meta’s more powerful model, Llama 4 Behemoth. Behemoth, which remains during active training, has so far shown significant advantages over established models such as GPT-4.5, Claude Sonnet 3.7 and Gemini 2.0 Pro, especially in the field of regular benchmarks. The insights and advanced methods from Behemoth have contributed to refining scouts and Maverick’s technical capabilities.
With the introduction of Llama 4, Meta AI Multimodal promotes artificial intelligence through highly refined and technically sophisticated models capable of deep semantic understanding and precise multimodal adaptation. This release exemplifies additional Meta AIS ongoing obligation to promote innovation and maintain open accessibility to researchers, developers and business applications.
Future progress in Multimodal AI is expected with the completion and public release of Llama 4 Behemoth. The original results indicate Behemoth’s potential to set new standards within multimodal performance, especially in Vote Applications and Calculation Reasoning Tasks. Meta AI plans to pass on detailed technical specifications and performance metrics after the end of the Behemoth model.
The message emphasizes Meta AIS dedication to pushing the technical limits of multimodal modeling, supporting the development of practical and research-oriented AI applications across various sectors, including scientific research, education and complex convertation systems. As Meta AI continues this course, the technological advances that are enshrined in Llama 4 scouts, Maverick and finally Behemoth are expected to facilitate significant progress in the computational and practical capabilities of Multimodal AI.
Check out Benchmarks and Download Llama 4. All credit for this research goes to the researchers in this project. You are also welcome to follow us on Twitter And don’t forget to join our 85k+ ml subbreddit.
🔥 [Register Now] Minicon Virtual Conference On Open Source AI: Free Registration + Certificate of Participation + 3 Hours Short Event (12 April, at [Sponsored]
Asif Razzaq is CEO of Marketchpost Media Inc. His latest endeavor is the launch of an artificial intelligence media platform, market post that stands out for its in -depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts over 2 million monthly views and illustrates its popularity among the audience.
