According to Google’s official announcement, Google officially released its most powerful cutting-edge AI model “Gemini 4 Carbon” on November 10. Compared with Argon, which focuses on daily complex tasks, Carbon is designed to overcome the most difficult long-term multi-step tasks and deep reasoning. It has real-time screen and program code debugging capabilities. It also launched the "/swarm" multi-agent collaborative architecture for the first time on the Antigravity platform, increasing the recall rate of open source security vulnerability mining by more than 25%. The experience is available to Ultra subscribers and global developers starting today. (Previous summary: Google released Gemini Robotics 2.5! It drives the Apollo 2 humanoid robot into the Mercedes-Benz production line with a "sorting success rate of 96%") (Background supplement: Google expanded the eighth generation TPU: adding 8tx and 8i Max, expected to be available in 2027) Technology giant Google once again ushered in an architectural breakthrough in the field of artificial intelligence (AI). Google officially announced on November 10 that its flagship cutting-edge model "Gemini 4 Carbon" will be fully available for use from now on. Officially defined as Google's most powerful model to date, it is specially built to solve deeply complex, ultra-long-horizon workflows (Long-horizon workflows) and difficult multi-step reasoning. It symbolizes that generative AI has officially entered a new stage of large-scale agents with deep autonomous execution capabilities. Designed for the toughest problems! Real-time on-screen debugging, say goodbye to "wild guessing" In terms of positioning division, Google has specifically clarified the complementary relationship between Carbon and the recently released Argon: Argon is good at handling most tedious software engineering and knowledge-based tasks, while Carbon is a trump card specially used to tackle "the most difficult core problems." In order to achieve in-depth exploration, Carbon uses a wider Token budget to conduct thorough logical reasoning and extended verification loops. On long-term projects, Carbon has demonstrated breakthrough self-correction capabilities. It can simultaneously write code and simultaneously compare the rendering results of the screen to troubleshoot errors. Whether it is building a complete application or generating a 3D virtual world, it can be adjusted independently. More importantly, when faced with missing or flawed underlying data, Carbon will not blindly speculate (illusion) like traditional large models, but will proactively flag warnings to users, greatly improving the rigor of the production environment. The first "/swarm" multi-agent collaboration: a single prompt word turns into a project army. The most important feature accompanying the debut of Carbon is the "/swarm" preview mechanism launched on the Antigravity platform. This feature completely overturns the past question-and-answer interaction model, allowing users to dynamically spawn a well-divided AI agent team in the background with just a prompt word, and supports real-time visual tracking throughout the process: High-level decision-making: Assign powerful Carbon to serve as the chief architect, workflow coordinator, and security auditor. Parallel execution layer: multiple parallel sub-agents in the background perform coding development, environment deployment and automated regression testing simultaneously. This team combat model has significantly shortened the development cycle of large-scale projects, further pushing the boundaries of automation in complex software engineering and cross-domain scientific reasoning. Information security detection rate soars by 25%! Ultra users are immediately available at the same price during the import period. According to data released by Google DeepMind Vice President Koray Kavukcuoglu on the official technology blog, Carbon has achieved leading results in multiple rigorous benchmark tests such as CursorBench, WANDR and AutomationBench. Especially in the field of network security and code review, Carbon's recall rate (Recall) in open source code vulnerability scanning has increased by more than 25% compared with the previous generation, and it has practical defense capabilities to independently verify and patch vulnerabilities. In terms of access rights, Google has adopted a comprehensive rollout strategy: End users: Google AI Ultra subscribers can call it directly in the Gemini App, the AI ​​mode in the search engine, the Antigravity platform, and Google AI Studio from now on. Developers and enterprises: The Gemini API can be called through Google AI Studio and Android Studio, and is also integrated into the Gemini for Business enterprise solution. In terms of billing strategy, Google has offered preferential measures and maintained the same rate as Argon in the initial introduction stage (US$2 per million input tokens and US$10 per million output tokens), which will be dynamically adjusted according to market conditions. With the launch of Gemini 4 Carbon, the focus of cutting-edge AI competition has shifted from pure parameter scale benchmarking to end-to-end delivery capabilities of multi-agent collaborative operations and real engineering tasks. Related Reports Google Guarantees TeraWulf’s $19 Billion Anthropic Transaction: Breakdown of Credit Support, Transaction Structure and Project Financing Risks Google Previews Institutional Encryption Custody Security System! MPC private key sharding is online, Alphabet has protected its ETH/SOL Discord and made 63,000 public servers searchable by Google! Quit teaching once and watch Google publish Gemini 4 Carbon! Leading all three information security benchmarks, winning the Intelligence Index with 59 points〈Google’s most powerful model, Gemini 4 Carbon, is online! Specializing in long-term reasoning, the information security vulnerability detection rate increased by 25%> This article was first published in Dongqu BlockTempo "Dongqu Dongzhi - the most influential blockchain news media".