Meta has launched Muse Spark, its new AI model, after investing nearly $15 billion in the Meta Superintelligence Lab, led by Alexandr Wang. This release marks the first tangible result of a strategic bet aimed at positioning Meta among industry leaders, directly competing with OpenAI, Google, and Anthropic. Muse Spark's strength lies in the quality of its data: in benchmarks where this aspect is crucial, such as multimodal vision or image comprehension, the model matches or surpasses its rivals. However, in areas like programming and abstract reasoning, it still lags behind industry leaders, indicating that Meta has optimized data-dependent aspects but has yet to master the more complex challenges of AI. Muse Spark integrates with Meta's ecosystem, allowing the use of content from Instagram, Facebook, and Threads, and anticipates incorporating posts in responses. It also introduces a "contemplation" mode, where multiple agents work in parallel, and more advanced models are already in development. Access is limited: users can try it, but the API is only available to selected partners. Open source has taken a back seat, similar to the situation in China: investments require monetization, and less information is shared with competitors. Anthropic, on the other hand, has launched Claude Mythos, a model focused on cybersecurity within the Glasswing project. In restricted tests, Claude Mythos has detected vulnerabilities that have been open for decades, such as flaws in OpenBSD, FFmpeg, and the Linux kernel. The goal is to use AI to strengthen global security, although access is restricted to selected organizations like AWS, Google, Microsoft, Apple, NVIDIA, and the Linux Foundation. This initiative, coordinated with governments and national security environments, reflects the growing geopolitical importance of cybersecurity. In the business sector, Anthropic has surpassed OpenAI in revenue, reaching $30 billion in annual recurring revenue by early 2026, driven by the adoption of its programming models, such as Claude, in nearly 1,000 companies that pay over a million dollars a year. The launch of Claude Managed Agents facilitates the creation of autonomous agents for business tasks, allowing companies like Notion, Rakuten, or Asana to automate processes in days. The new GLM-5.1 model stands out for its persistence and adaptability, autonomously solving complex problems and iteratively improving results. OpenAI has published a strategic document outlining how AI could transform the economy, suggesting shifting taxes from labor to capital and considering AI as basic infrastructure. It proposes measures such as taxes on robots, public wealth funds, and a four-day workweek, along with massive investments in chips, data centers, and energy. In research, Google has introduced PaperOrchestra, a multi-agent system that automates the writing of scientific articles from disorganized materials, outperforming other solutions in synthesis quality and structure. MIT has developed SEAL, an approach that allows language models to adapt and learn autonomously, generating and optimizing their own training data, which could revolutionize AI development. In robotics, initiatives include robots that clean beaches, pesticide-free ecological crops thanks to robots that remove weeds, and a robot marathon in China. In the business sector, Eric Boyd (Microsoft) joins Anthropic as head of infrastructure, Intel collaborates with SpaceX and Tesla on the Terafab project, and Elon Musk attempts to influence OpenAI's governance. Intel focuses on chip packaging, OpenAI tests its image generation model Image V2, Google develops the Jules V2 agent for programming, and Anthropic signs energy agreements with Google and Broadcom. Meta will maintain some open-source models, but only the smaller ones in the future, marking a shift similar to that of China.