1. Google DeepMind releases Gemini 3.1 Pro, ARC-AGI-2 score reaches 77.1%
Google DeepMind launched Gemini 3.1 Pro, currently the most powerful Pro-level model, which supports ultra-long context windows of up to 1 million tokens and achieved a high score of 77.1% in the extremely challenging ARC-AGI-2 benchmark. The model has multi-modal reasoning capabilities across text, images, audio, video and code, marking an important step in the multi-modal performance of the Gemini series. Meanwhile, Gemini version 3.0 is expected to be released in the first quarter of 2026 and will support real-time 60fps video processing and built-in inference capabilities.
2. China released five large models in March, with the MiniMax M2.5 catching up with the Claude Opus 4.6
China's AI industry is experiencing a wave of intensive releases this month, with Tencent, Alibaba, Baidu, ByteDance and MiniMax launching new models one after another. Among them, MiniMax’s M2.5 model performs on par with Anthropic’s Claude Opus 4.6 in multiple benchmark tests, but its cost is significantly lower than the latter, which has attracted widespread attention in the industry. This reflects the rapid rise of China's large model manufacturers in cost-effective competition, and also brings greater price pressure to overseas head models.
3. Anthropic Claude Opus 4.6 introduces effort level control and context compression mechanism
Anthropic has launched two important new features for the flagship model Claude Opus 4.6: Effort Controls allow developers to flexibly weigh intelligence level, speed and cost among four levels (low, medium, high, and maximum); Context Compaction significantly extends task execution time without touching the token limit by automatically summarizing historical conversations. These features are particularly suitable for long-running agent tasks, further strengthening Claude's usefulness in agentic scenarios.
4. Samsung cooperates with Google to install Gemini AI on 800 million mobile devices by the end of the year
Samsung released the Galaxy S26 series of mobile phones, equipped with AI functions that can predict user intentions and complete tasks autonomously. At the same time, Samsung Electronics announced a deepening cooperation with Google and plans to double the number of mobile devices equipped with Gemini AI to 800 million units by the end of 2026, covering mid- to low-end models and accelerating the popularization of generative AI to the masses. This is a major leap forward in the scale of mobile AI deployment.
5. Cursor releases Agent Trace draft specification to establish attribution standards for AI-generated code
Code editor Cursor has released a draft of the Agent Trace open specification, which aims to establish a unified standard for attribution records of AI-generated code in version control systems. The specification adopts a vendor-neutral format, which can clearly mark in tools such as git which code is written by AI and which is completed by humans, which will help make future compliance audits and collaboration transparent. Vercel released the open source tool Skills.sh at the same time, providing a standardized command line reusable action interface for AI agents.
6. The battle between the U.S. federal government and states over AI regulation is heating up, with a key node coming on March 11
The Trump administration signed an executive order in December 2025 calling for a unified federal AI regulatory framework and used the withholding of $21 billion in broadband funding as leverage to suppress state legislation. The executive order stipulates that a list of "excessive" state-level AI laws must be published before March 11, 2026, and the Department of Justice simultaneously established an AI litigation task force. AI regulations in states such as Colorado, California, and Texas will still take effect as scheduled, and the game of regulatory power between the federal and state governments is expected to continue to intensify.
7. Anthropic donates $20 million to political groups, betting on AI regulation ahead of election
Anthropic announced a $20 million donation to Public First Action, a political group that advocates for AI regulation, to support bipartisan candidates and actively intervene in policy trends ahead of the 2026 midterm elections. The move is rare in the industry, as the Trump administration has been critical of agencies advocating for AI safety regulation. Analysts believe that Anthropic aims to seize the right to speak in advance for possible future regulatory frameworks.
8. China’s new five-year plan lists quantum computing and AI as core strategic goals
China's latest five-year plan clearly lists quantum computing leadership and the expansion of artificial intelligence as national strategic priorities, emphasizing increasing investment in basic research, improving computing infrastructure, and promoting the deep integration of AI with real economies such as manufacturing, medical care, and finance. This strategic layout has formed a direct competition with the United States' "AI dominance" strategy, and the international AI geopolitical game has further intensified.
// KEY TAKEAWAYS
This issue of news presents three main lines: first, model competition is intensifying. Google, Anthropic, and many Chinese manufacturers have intensively released new models. The price war between MiniMax M2.5 and Claude Opus 4.6 has officially begun; second, AI From large and comprehensive to practical, efforts such as hierarchical control, context compression, agent standard specifications and other functions mark the new focus of implementation; thirdly, the regulatory and geographical landscape is accelerating differentiation, the legislative game between the US federal and state governments, Anthropic's political bets, and the head-on competition between China and the United States at the AI strategic level, indicating that the direction of AI in 2026 will be increasingly influenced by policy forces.