Llama 4 Scout
Llama 4 Scout is an open-weight mixture-of-experts model from Meta, released on April 5, 2025, with 17B active and 109B total parameters. It takes multilingual text and image input and returns text and code.
Released
April 5, 2025
Type
multimodal
License
open-weight
Llama 4 Community License Agreement
Context
10,000,000 tokens
Listed as 10M tokens on Meta’s model card. Meta says the model was trained and post-trained at 256K.
Capabilities
Architecture
Parameters: 109B total, 17B active (mixture of experts, 16 experts)
Sources
Every figure on this page was checked against these pages on . A figure we could not source is left out.
- Meta AI blog: Llama 4 (opens in a new tab)
- Model card: Llama-4-Maverick-17B-128E-Instruct (covers Scout and Maverick) (opens in a new tab)
Spotted something out of date? Tell us.
Links
In the news: Llama 4 Scout and Meta
The Decoder
Meta's new frontier model Muse Wonder Full breaks into the AI top tier
Meta's new Muse Wonder Full frontier model has entered the AI top tier, achieving a top-three ranking on six out of ten benchmarks. The models demonstrated the biggest gains in agentic tasks. However, the cost per task has nearly tripled. The article "Meta's new frontier model Muse Wonder Full breaks into the AI top tier" appeared first on The Decoder.
The Hacker News
Bring Your Own Agent: Your Employees' Assistants Are Already Logged In
The rollout of personal AI agents is encountering significant challenges, including privacy concerns and identification issues. Amazon has blocked Meta's Muse agent for failing to identify itself as a bot during browsing, and the Instinct agent has also raised privacy alarms. Despite these ongoing issues and accumulating negative experiences, the technology sector remains captivated by the capabilities and potential of these personal AI services. The article highlights that the personal AI agent rollout is proceeding with expected difficulties.
Fast Company
Data centers produce heat year-round. Here’s how to save it for winter
Epic Systems, a health software company in Verona, Wisconsin, has implemented a geothermal system to manage waste heat from its 3.5 megawatt (MW) data center. Over two decades, the company has drilled approximately 6,100 wells, ranging from 300 to 800 feet deep, to heat and cool about 40 buildings on its campus. The system's unique aspect is its ability to store heat in the ground during summer and retrieve it months later for winter use. This addresses a common challenge in data center heat reuse, as servers generate heat year-round, while buildings typically require heating for only about four months of the year. Unlike ordinary ground-source heat pumps that extract heat from the earth in winter, Epic's system injects heat into the ground during warmer months. This approach is distinct from other recent news involving data centers and geothermal energy, such as Google's purchase of geothermal power in Nevada and Meta's contracts in New Mexico. Those projects focus on extracting existing heat from the earth to generate electricity. Epic's method, however, involves storing heat generated by the building itself underground, with the ground acting as a thermal reservoir. The concept highlights the importance of storage for effectively reusing data center waste heat, drawing a parallel to how large snow piles can persist long after smaller ones have melted.
Fast Company
Agentic AI just got very real
Agentic AI has transitioned from a theoretical concept to a tangible workflow enhancement, as exemplified by OpenAI's new Dots agent. The agent recently proved its practical utility by proactively identifying and resolving a significant scheduling conflict for a user. Dots alerted the user to a discrepancy between a speaking engagement entry in their calendar and a confirmation email, which indicated a different date for the event. This proactive intervention by Dots prevented the user from attending an event on the wrong day, a scenario that would have resulted in an empty room for an audience. OpenAI announced Dots on September 29 at its DevDay conference keynote in San Francisco. Currently, access to Dots is limited to users subscribed to OpenAI's premium ChatGPT tiers, with pricing starting at $100 per month. OpenAI intends to make agentic AI broadly available to a mass market, potentially for billions of people, once Dots is further refined and its economic model is established. The service operates on OpenAI's GPT-6 Astra model, announced earlier in September, and provides each user with a dedicated Linux computer in the cloud for the Dots agent to utilize. CEO Sam Altman expressed the company's long-term vision for agentic AI during a post-keynote press Q&A, stating, "You should of course expect us to do a mass-market thing for billions of people someday." Even prior to the debut of Dots, personal AI agents were gaining significant traction in the market. Meta's free Muse agent has achieved the top position on Apple's list of most-downloaded iPhone applications. Additionally, Instinct, a startup, offers its namesake agent that operates within popular texting applications such as WhatsApp and Apple Messages. Although Instinct's agent currently has a waitlist, it has garnered substantial attention and a valuation of $10 billion within Silicon Valley, highlighting the growing interest and investment in agentic AI technologies. The ability of these agents to autonomously manage tasks and provide proactive assistance is reshaping user interaction with AI, moving towards more integrated and supportive digital assistants. The practical demonstration by Dots underscores the growing potential for AI agents to become indispensable tools for managing daily tasks and improving personal productivity.
Decrypt
AI Startup Manus Raises $500 Million After China Nixed Meta’s $2 Billion Acquisition
AI startup Manus has successfully raised over $500 million in fresh funding, a substantial financial injection that follows the unexpected termination of its previously agreed-upon acquisition by Meta Platforms. The deal, which was reportedly valued at approximately $2 billion, was ultimately blocked by Chinese regulators, preventing Meta, a global technology giant known for its social media platforms like Facebook and Instagram and its significant investments in virtual and augmented reality through its Reality Labs division, from integrating Manus's advanced AI capabilities. This new funding round signifies a strong vote of confidence in Manus's technology and its independent future. Manus's core innovation lies in its development of a self-driving AI assistant. This technology aims to create artificial intelligence systems capable of operating autonomously, understanding complex environments, and executing tasks with a high degree of independence, extending beyond the realm of autonomous vehicles to a broader spectrum of applications. The company's early focus on this sophisticated AI approach predates the current widespread surge in generative AI and large language models, positioning Manus as a potential pioneer in the field of advanced AI agents. The concept of AI assistants that can navigate and act within digital or physical spaces autonomously is a key frontier in AI research, with implications for numerous industries. The initial acquisition by Meta was seen as a strategic move to bolster Meta's own AI ambitions, particularly in developing more sophisticated virtual assistants and enhancing its metaverse initiatives. Meta, under CEO Mark Zuckerberg, has been heavily investing in AI research and development to power its future products and services. However, the proposed acquisition faced significant hurdles with Chinese authorities, who ultimately exercised their regulatory power to halt the transaction. This regulatory intervention by Beijing is not uncommon in high-stakes technology deals involving foreign entities and China, often citing concerns related to national security, data privacy, or market competition. Following the collapse of the Meta deal, Manus was compelled to seek alternative avenues for growth and capital. The successful closing of a funding round exceeding $500 million, though the specific investors remain undisclosed, demonstrates the market's belief in Manus's technological prowess and its potential to disrupt various sectors. This capital infusion will be crucial for Manus to accelerate its research and development, expand its engineering talent, and further commercialize its self-driving AI assistant technology in a highly competitive AI landscape. The AI sector is currently experiencing unprecedented growth and investment, with numerous startups vying for market share and talent, making Manus's achievement particularly noteworthy.