A Guth Labs publication ↗
Back to News

Google opens Gemini 4 Carbon to paid users and previews agent teams

AI-generated conceptual editorial illustration of graphite AI core coordinating abstract coding workspaces; branded guthlabs.ai.
AI-generated conceptual editorial illustration | guthlabs.ai

The wider rollout brings an Antigravity /swarm preview and introductory API prices that double after November 30.

Google opened Gemini 4 Carbon to paying developers on November 10, bringing its latest model into the Gemini API through AI Studio and Android Studio. The company is pitching Carbon for difficult coding, research and other tasks that require sustained work across multiple steps. Google developer announcement

The wider release also reaches Gemini for business, while Google AI Ultra subscribers are getting access in the Gemini app, AI Mode in Search and Antigravity. This is the public rollout of a model first announced on October 28 with access limited to selected cybersecurity defenders. Google rollout announcement October announcement

API pricing starts at $2 per million input tokens and $10 per million output tokens. Google says the introductory period expires on November 30, after which the rates become $4 and $20 respectively. Teams evaluating Carbon now therefore need to budget for the higher rates as well as the introductory offer. Google rollout announcement

Alongside the model, Google is previewing /swarm in the Antigravity app and command-line tool. The feature divides a project among a coordinator, specialist workers and agents assigned to test and challenge changes. Workers edit in separate Git worktrees and communicate through a shared message board. A live interface lets users inspect activity and steer individual agents. Antigravity announcement

The design lets teams assign different models to different roles. Google suggests using Carbon for demanding coordination, architecture and verification, with lighter models handling routine implementation. Its announcement also warns that parallel agent teams can consume usage quota much faster than a normal chat, even when idle agents use no tokens. Antigravity announcement

That warning is central to evaluating the launch. A lower token price does not necessarily produce a lower bill for completed work: a longer reasoning run, more parallel agents or repeated revisions can consume the apparent saving. A useful comparison holds the required result and review standard constant, then measures the complete workflow rather than one request.

Google’s model card lists a combined input-and-output context window of up to one million tokens, with text, image, audio and video inputs and text output. It also acknowledges hallucinations, occasional slow responses or timeouts, and increased token use at higher thinking settings. Specialized cybersecurity capabilities remain available to verified defenders through Fairwind; the model card says its cybersecurity benchmark results reflect that access route. Model card

For software teams, the practical test will be the cost of a reviewed, working change. Parallel workers can shorten elapsed time, but their output still has to fit together and survive testing. Comparing Carbon against an existing workflow should include retries, human review, usage limits and the December prices. The release gives more customers a chance to run that comparison; the launch claims alone cannot settle it.

Sources and citations

Revision history

Published .