Meta Releases Llama 4 Scout and Maverick: Its Most Advanced Multimodal AI Models Yet
Meta Platforms has announced the launch of its latest large language models (LLMs), Llama 4 Scout and Llama 4 Maverick, designed to lead in the field of multimodal artificial intelligence.
Llama 4 Models Introduced
On Saturday, Meta Platforms Inc. released two new versions of its Llama large language model:
- Llama 4 Scout
- Llama 4 Maverick
Both models are being positioned as the company’s most technically advanced AI systems to date. According to Meta, these models excel in multimodal capabilities, meaning they can process and integrate data across multiple formats, including text, video, images, and audio.
“These are the best in their class for multimodality,” Meta stated in its press release.
Open Source Commitment
- Meta confirmed that both models—Scout and Maverick—will be released as open source software.
- This move is aimed at encouraging broader adoption and innovation across the AI development community.
Llama 4 Behemoth Previewed
- Alongside the main launch, Meta also previewed a third model named Llama 4 Behemoth.
- Described as “one of the smartest LLMs in the world,” Behemoth is expected to act as a training teacher model for the new AI systems.
Development Delays and Performance Concerns
- According to a report by The Information, the launch of Llama 4 was previously delayed.
- Internal testing revealed that Llama 4 did not meet Meta’s performance benchmarks, particularly in mathematical reasoning and logic tasks.
- Concerns were also raised about its inferior performance in humanlike voice conversation, compared to competing models such as OpenAI’s GPT series.
$65 Billion AI Investment Plan
- In response to growing competition and investor pressure, Meta plans to invest up to $65 billion in 2025 in AI infrastructure.
- The investment is part of a larger trend among tech giants racing to enhance their machine learning capabilities following the global success of OpenAI’s ChatGPT.
