Google made Gemini 4 Carbon generally available yesterday. Two weeks ago the model went only to cybersecurity defenders through the Fairwind Program. Now it reaches all paying developers and enterprises through the API in Google AI Studio and Gemini for business, and it is rolling out to every Google AI Ultra subscriber — in Google Antigravity, the Gemini app, and AI Mode in Search.
The price is the surprise
Carbon costs the same as Gemini 4 Argon: $2 per million input tokens and $10 per million output tokens, as an introductory price. Carbon gets expensive only because it burns more tokens — it explores longer, reasons deeper, and runs extended verification loops.
Koray Kavukcuoglu, SVP at Google DeepMind, is clear about the split. Argon stays the right choice for the majority of software engineering and knowledge work. Carbon is built for tasks that test the limits of a model’s intelligence: complex multi-step coding and debugging, open-ended scientific research.
The benchmarks come with a cost figure
Google claims first place 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. There is also a new benchmark, built with Artificial Analysis: Multimodal Knowledge Agent, or MKA. It tests agentic work over dense files with charts, images, blueprints, schematics, and tables. Carbon ranks second there, at more than 6x lower average cost per task than the top performer.
That Google refuses to quote the rankings without the cost figure is the real argument for this generation of models.
/swarm in Antigravity
For larger projects, Google is previewing /swarm in Google Antigravity: a single prompt turns into a coordinated team of agents you can watch in real time. Carbon takes the roles that need the deepest reasoning — architectural design, coordination, auditing — while faster, lighter models run across parallel isolated workspaces and adversarial verifiers attack every change. As an example, a team of Carbon agents turned historical photos and engineering plates into an interactive 3D exhibition of the Eiffel Tower’s 1887–1889 construction, from one prompt.
Same week, same shape
The architecture is the one Anthropic showed for Fable 5.2 and Google itself showed for AlphaProof Code: one strong model doing architecture and audit, many smaller ones on the assembly line, and verifiers paid to attack the result. Three labs, one week, the same answer to the question of what an expensive model is for. Cost per task rather than cost per token is the accounting that follows from it.