Gemini 4 Carbon is now open to developers, enterprises, researchers and everyday users. Carbon is Google’s newest frontier model, delivering leading performance on cybersecurity, long‑horizon agentic workflows, complex coding and scientific reasoning. Two weeks ago it was released to security teams through the Fairwind program, and it now extends the capabilities of Argon.\n\nOn independent benchmarks such as CursorBench 4.0, Perplexity WANDR and Zapier AutomationBench, Carbon ranks first while costing more than two times less per task. Using the /swarm feature in Google Antigravity, a single prompt can spawn a coordinated team of agents that can be tracked in real time. For example, /swarm turned historic photos and engineering plates into a Three.js 3D exhibition, and built a 906K‑triangle office model within hours.\n\nIn high‑precision knowledge work across legal, finance and enterprise operations, Carbon combines visual understanding and exhaustive retrieval to produce polished spreadsheets, presentations and reports that follow strict schemas. To measure multimodal agent performance, Google partnered with Artificial Analysis to launch the Multimodal Knowledge Agent benchmark, where Carbon placed second while costing six times less than the top model.\n\nFor software engineering and 3D UI design, Carbon excels in multi‑agent collaboration, handling cross‑subsystem root‑cause analysis, build‑and‑verify loops and delivering complete web applications. With /swarm, Carbon created an open‑world 3D game where a pelican rides a GBike to explore Google’s Mountain View campus, and generated new art direction concepts and product imagery for the Google Merchandise Store.\n\nIn cybersecurity, Carbon leads on benchmarks such as SEC‑bench Pro, ExploitGym and CWE‑bench v1, detecting, exploiting and generating verified patches. In the Fairwind program Carbon achieved 100% recall on scanning and 95% precision on triage. External tests showed it can surface previously hidden 0‑day vulnerabilities within a few hours.\n\nScientific reasoning sees strong gains across physics, math and life sciences. Carbon helped Axiom Bio discover novel toxicity mechanisms, and powered the AlphaProof Code project to generate millions of lines of Lean code for formal verification of the Signal cryptographic protocol, cutting verification time from years to weeks. In a multi‑agent search, Carbon increased homolog retrieval for low‑homology protein families by roughly 40‑fold, feeding richer data to AlphaFold‑3 style structure predictors.\n\nSafety and alignment improvements include a 99% block rate on simulated attack trajectories, reducing per‑attempt breach probability to 0.8%. On a 375‑task agentic honesty benchmark Carbon passed 92% of tasks and achieved a 0% success rate on the Gray Swan IPI prompt‑injection benchmark.\n\nPricing and access: Carbon is available via the Gemini API paid tier in AI Studio, Gemini for Business and Google Antigravity at $2 per million input tokens and $10 per million output tokens—the same introductory price as Argon. Enterprises and developers can use it in paid plans, Google AI Ultra subscribers can try it in the Gemini app and Search AI Mode. Researchers may apply for free six‑month access to both Argon and Carbon through Antigravity’s highest‑usage tier, and security defenders can join the Fairwind program.\n\nReview: Carbon bridges cutting‑edge model capabilities with practical economics, offering a viable solution for multimodal, agentic and security‑critical workloads.