Google said it is making Gemini 4 Carbon available to developers, enterprises, and scientists as of Nov. 10.[1] The model is accessible via the API in Google AI Studio and Gemini for business, and for Google AI Ultra subscribers in Google Antigravity, the Gemini app, and AI Mode in Search.[1]

Carbon uses more tokens than Argon to explore options thoroughly, reason deeply, and run extended verification loops, Google said.[1][2]

On independent benchmarks, Carbon ranks #1 on CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench at more than 2X lower cost per task compared to the next best external model, Google reported.[1][2] It also ranks #2 on the new Multimodal Knowledge Agent (MKA) benchmark, which tests a model’s multimodal performance in agentic workflows over dense files with charts, images, blueprints, schematics, and tables, while being more than 6X lower in average cost per task than the top-performing model, Google said.[1] Carbon scored a 0% prompt-injection attack success rate on the Gray Swan IPI benchmark test, CNET reported.[3]

To support complex projects, Google is previewing /swarm in Google Antigravity, a feature that turns a single prompt into a coordinated team of agents, allowing developers to assign Carbon to roles requiring deep reasoning while pairing it with faster, lighter models in parallel workspaces, the company said.[1] Carbon has already helped uncover zero-day vulnerabilities in under two hours and advanced drug toxicity research by identifying new toxicity mechanisms, Digital Trends reported.[2]