Google announced that Gemini 4 Carbon, its strongest frontier model, is now available to developers, enterprises, scientists, and consumers, ending a two-week exclusive window in which the model was restricted to a small group of cybersecurity defenders. According to Google's official announcement, Carbon is rolling out through the API in Google AI Studio, to business customers, and to all Google AI Ultra subscribers across Google Antigravity, the Gemini app, and AI Mode in Search.
The general availability marks an unusually fast transition from controlled release to open access, and it arrives alongside a significant new capability: a preview of /swarm in Google Antigravity, a feature that turns a single prompt into a coordinated team of AI agents. For a rundown of how Carbon compares with the rest of Google's frontier lineup, see our latest AI news coverage.
From Cybersecurity Defenders to Everyone
Two weeks ago, Google released Carbon to a selected group of cybersecurity defenders through its Fairwind Program, a staged rollout strategy the company has increasingly favored for models with unusual capability profiles. Carbon is designed for frontier performance in cybersecurity capabilities and long-horizon agentic workflows spanning complex knowledge work, software engineering, and scientific reasoning.
In the announcement, Google drew a clear line between Carbon and its sibling model Gemini 4 Argon, which remains the company's recommendation for the majority of software engineering and knowledge work. Carbon, by contrast, is positioned for tasks that test the limits of a model's intelligence: complex, multi-step coding and debugging, and open-ended scientific research and discovery.
Benchmark Claims and Argon-Level Pricing
Google backed the release with a set of third-party benchmark results. Carbon ranks first on CursorBench 4.0, Perplexity's WANDR, and Zapier's AutomationBench, in each case at more than 2X lower cost per task compared to the next best external model, according to the announcement.
The company also partnered with evaluation firm Artificial Analysis to launch a new Multimodal Knowledge Agent (MKA) benchmark, which tests how models reason natively over dense files containing charts, images, blueprints, schematics, and tables. Carbon ranks second on the new benchmark while costing more than 6X less per task on average than the top-performing model.
Pricing follows Google's recent pattern of aggressive rate cards for frontier models: Carbon is available at the same introductory price as Gemini 4 Argon, at $2 per million input tokens and $10 per million output tokens, for all paying developers and enterprises via the API in Google AI Studio and Gemini for business.
/swarm Turns One Prompt Into an Agent Team
The most distinctive addition shipping alongside general availability is /swarm, now in preview in Google Antigravity. The feature converts a single prompt into a coordinated team of agents that developers can track in real time. Google's suggested pattern assigns Carbon to the roles that demand the deepest reasoning, such as architectural design, coordination, and auditing, while faster, lighter models work across parallel isolated workspaces. Adversarial verifiers then test each other's output.
To illustrate the approach, Google described two demos built with Carbon teams. In one, a swarm of agents turned historical photographs and engineering plates into an interactive Three.js 3D exhibition of the Eiffel Tower's 1887–1889 construction from a single prompt. In another, a multi-agent team processed seven public city council planning PDFs, extracted 206 engineering drawing views, and produced an interactive 906,000-triangle Three.js reconstruction of Google's Platform 37 office across ten overnight loops.
Positioning for Enterprise and Finance Work
Google is also pitching Carbon at document-heavy enterprise domains. According to the announcement, the model combines frontier visual understanding, reasoning, and exhaustive retrieval to synthesize grounded insights across large sets of dense documents, and follows strict schemas and formatting to generate polished spreadsheets, presentations, and reports.
The financial sector features prominently in Google's claims. In pre-release evaluations, Google says Carbon matched the performance of the best model tested on deep financial research in Hebbia's E2E Financial Agentic evaluation, and ranked at the top out of 23 frontier model configurations on Balyasny Asset Management's General Financial Intelligence benchmark, which spans legal expertise, agentic tool use and execution, coding and data engineering, and perception and context understanding.
A Rapid Frontier Cadence
The release continues a striking cadence for Google: Gemini 4 Argon launched first, and Carbon arrived as the follow-up frontier model roughly two weeks after its own controlled debut. The company argues the two models now serve distinct tiers of demand, with Argon handling everyday engineering work at scale and Carbon reserved for the hardest agentic and scientific workloads.
With consumer access through the Gemini app and AI Mode in Search, developers gain API access at Argon pricing, and the /swarm preview pointing toward multi-agent workflows as a first-class product surface, Google is making a clear bet that frontier intelligence can be delivered both broadly and economically. Whether the benchmark economics hold up in production workloads will be the next test for a model the company calls its strongest yet.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
