Google has opened up its most powerful AI model to a much wider audience. On Tuesday, November 10, the company made Gemini 4 Carbon available to developers, businesses, scientists and some consumers, describing it as its strongest model for complex coding, long-running agent tasks, scientific reasoning and cybersecurity.

Carbon had been available for about two weeks to a limited set of cybersecurity defenders through Google’s Fairwind Program. The broader launch was announced in a post on Google’s official blog by Koray Kavukcuoglu, SVP of Google DeepMind and the company’s chief AI architect.

Carbon vs. Argon: which model is for what

Google now offers two frontier Gemini 4 models. According to the company, Gemini 4 Argon remains the best choice for most software engineering and knowledge work. Carbon is aimed at the hardest jobs, such as complex multi-step coding and debugging and open-ended scientific research. It uses more tokens than Argon to explore problems thoroughly, reason in depth and run extended checks on its own work.

Google says Carbon ranks first on several third-party benchmarks built around real workflows, including CursorBench 4.0, Perplexity’s WANDR and Zapier’s AutomationBench, at less than half the cost per task of the next-best external model. As always with vendor-reported benchmarks, real-world results will vary by task.

Swarms of agents

Alongside Carbon, Google is previewing a feature called /swarm in its Google Antigravity development tool. It turns a single prompt into a coordinated team of AI agents that can be tracked in real time. Developers can assign Carbon to roles needing the deepest reasoning, such as architecture, coordination and auditing, while pairing it with faster, lighter models working in parallel.

Google’s demos include an interactive 3D reconstruction of the Eiffel Tower’s construction built from historical photos, and a browser game created by a team of Carbon-led agents.

A big focus on cybersecurity

Carbon’s cyber capabilities are a major theme. Through the Fairwind Program, Google says the model leads benchmarks for finding, exploiting and patching software vulnerabilities. In Google’s own internal testing, the company reports Carbon hit 100 percent recall on scanning and 95 percent precision on triage in its Project Fortify system. Security firm Wiz, Google says, used Carbon to find previously unknown vulnerabilities in two major SaaS providers in under two hours and worked with them to fix the issues.

Because stronger cyber skills could be misused, Google says it tightened safeguards. In simulated attack testing, it reports Carbon blocked 99 percent of attack attempts from an adaptive red-teaming agent and lowered the per-attempt breach rate to 0.8 percent, down from 3.1 percent for Argon. Access to the full “Carbon for cyber” capabilities remains limited to approved defenders.

Science and research uses

Google also highlighted research applications. Its teams used Carbon to help with formal verification of the code behind Signal’s cryptographic protocol, to diagnose subtle lab experiment errors and to find far more related protein sequences for structure prediction. Google is expanding a program that gives selected scientists free access to Argon and Carbon for six months, subject to eligibility and capacity.

Price and availability

Developers can use Carbon through the Gemini API in Google AI Studio, and businesses through Gemini for business. Introductory pricing matches Argon, at $2 per million input tokens and $10 per million output tokens, until November 30, 2026, after which it rises to $4 and $20. For consumers, Carbon is rolling out to Google AI Ultra subscribers in the Gemini app and AI Mode in Search. Google says the model card and an updated developer guide explain how to migrate existing workflows.

The bottom line

Gemini 4 Carbon is Google’s bid to lead on the hardest, most expensive AI work, from large coding projects to security research. Most everyday users will not touch it directly unless they pay for Google’s top tier, but the tools it powers, especially agent swarms and automated security scanning, show where AI products are heading next.

Image: Free computer server room image via Rawpixel, CC0 1.0