Cursor puts high effort first at 60.4 percent

The Cursor account posted at 16:35 UTC on 10 November 2026, which is 00:35 in Beijing on 11 November. The post says Gemini 4 Carbon is now available in Cursor, and that at high effort it is the top model on CursorBench at 60.4 percent. The Google blog carries only the date 2026-11-10 and no clock. This column times the availability claim from the Cursor post, and does not turn the date-only blog into a minute.

The blog says developers, enterprises, scientists, and consumers can now try the model. It calls Carbon the strongest model, with frontier performance in cybersecurity capabilities and long-horizon agentic workflows across complex knowledge work, software engineering, and scientific reasoning. Two weeks earlier, the blog says, Carbon was released to a set of cybersecurity defenders through the Fairwind program. Argon remains, in the blog's wording, the choice for most software engineering and knowledge work. Carbon is described as the model for harder tasks, including multi-step coding and debugging and open-ended scientific research, and as using more tokens than Argon. Cybersecurity here is the capability the blog names. This column does not describe how vulnerabilities are found or exploited.

On independent benchmarks of real workflows, the blog says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR, and Zapier's AutomationBench, at more than 2X lower cost per task than the next-best external model. The 60.4 percent figure is only in the Cursor post. The blog does not print that percentage. The post says the score is at high effort. The blog says it launched the Multimodal Knowledge Agent benchmark, MKA, with Artificial Analysis. On that test the blog ranks Carbon second, at more than 6X lower average cost per task than the top model. It does not name that top model.

The blog prices public Carbon at the same introductory rate as Argon: 2 dollars per million input tokens and 10 dollars per million output tokens. A footnote on the same page says the introductory period ends on 30 November 2026, after which the price is 4 dollars per million input tokens and 20 dollars per million output tokens, and that this applies only to the Carbon model intended for public access. Paying developers and enterprises can use the API in Google AI Studio and Gemini for business. Google AI Ultra subscribers also get it in Google Antigravity, the Gemini app, and AI Mode in Search.

The blog says that four weeks earlier it opened applications for 10,000 scientists to use Argon in Google Antigravity free of charge for six months, and that thousands have already been onboarded. The 10 November post says access is expanding to Carbon for current and future participants, and that the program will keep offering the latest frontier models. Applications stay open, with weekly onboarding, subject to eligibility, capacity, and selection. The blog also previews /swarm in Antigravity, which turns one prompt into a team of agents that can be watched together. One example uses seven public planning PDFs, 206 extracted engineering-drawing views, and a single prompt. The blog says a Carbon team built an interactive reconstruction of Google's Platform 37 office across 10 overnight loops, with 906K triangles.

The photograph is the author portrait on this blog, Koray Kavukcuoglu, identified there as senior vice president at Google DeepMind and chief AI architect at Google. The original file is square. A center crop to 4:3 removes a strip above the head and along the bottom edge, and leaves the face in the middle. It is not a photograph of a launch event.