Google makes Gemini 4 Carbon available to developers and consumers

Availability and pricing

Gemini 4 Carbon is available through the API in Google AI Studio and Gemini for business for all paying developers and enterprises. Google DeepMind says it has the same introductory price as Gemini 4 Argon, $2 per million input tokens and $10 per million output tokens.

The model is also rolling out to all Google AI Ultra subscribers in Google Antigravity, the Gemini app and AI Mode in Search. Google DeepMind previously released Carbon to a group of cybersecurity defenders through its Fairwind Program two weeks before this wider availability.

According to Google DeepMind, Argon remains the preferred option for most software engineering and knowledge work, while Carbon is designed for demanding tasks such as complex, multi-step coding and debugging, and open-ended scientific research. The company says Carbon uses more tokens than Argon to explore problems thoroughly, reason deeply and run extended verification loops.

A model aimed at complex knowledge work

Google DeepMind says Carbon combines visual understanding, reasoning and retrieval to work across dense documents and data in areas including legal work, finance and enterprise operations. It can follow strict schemas and formatting to produce spreadsheets, presentations and reports.

The company partnered with Artificial Analysis on the Multimodal Knowledge Agent benchmark, which tests multimodal performance in agentic workflows involving files with charts, images, blueprints, schematics and tables. Google DeepMind says Carbon ranked second on the benchmark while costing more than 6X less per task on average than the top-performing model.

In other company-reported results, Carbon ranked 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. Google DeepMind also says Carbon matched the best model tested in Hebbia’s E2E Financial Agentic evaluation and ranked first among 23 frontier model configurations on Balyasny Asset Management’s General Financial Intelligence benchmark.

Cybersecurity, coding and agent teams

Carbon is also being positioned as a cybersecurity model. Through the Fairwind Program, Google DeepMind says it leads on benchmarks covering vulnerability discovery, exploitation and patching. In Project Fortify, the company reports 100% recall for scanning and 95% precision for triage on an internal benchmark.

Google DeepMind says testing on open-source code repositories showed an over 25% relative improvement over Argon’s recall rate, including the discovery of previously undetected real-world logic flaws. Its Cloud Vulnerability Research team used Carbon to show that two bugs in open-source software were exploitable vulnerabilities, leading maintainers to reclassify them as critical and issue a patch. Details will be shared after responsible disclosure is complete.

Google is also previewing /swarm in Google Antigravity. The feature turns one prompt into a coordinated team of agents that users can track in real time. Developers can assign Carbon to roles requiring deeper reasoning, such as architectural design, coordination and auditing, while using faster models in parallel workspaces and adversarial verifiers. The capability extends Carbon’s intended use beyond individual answers to larger coding, design, research and software projects.