Google rolls out Gemini 4 Carbon to AI Ultra subscribers at $2/$10 per 1M tokens, Argon stays

Google is rolling out Gemini 4 Carbon to AI Ultra subscribers in the Gemini app, AI Mode and Antigravity, following a preview at the end of October, according to 9to5Google. Introductory API pricing is $2 per 1M input tokens and $10 per 1M output tokens until November 30. Afterward, it will cost $4 per 1M input tokens and $20 per 1M output tokens.

Carbon and Argon side by side

Argon, available to AI Pro subscribers and up, remains in the Gemini app alongside Carbon. Google positions the two differently: "Argon remains the ideal choice for the majority of software engineering and knowledge work, while Carbon is built for tasks that test the limits of a model's intelligence." Google's examples for Carbon include complex, multi-step coding and debugging, and open-ended scientific research and discovery.

Antigravity /swarm

Google Antigravity is previewing /swarm, which turns a single prompt into a coordinated team of agents that users can track in real time. Developers can assign Carbon to roles that demand the deepest reasoning, such as architectural design, coordination and auditing. Faster, lighter models handle work in parallel isolated workspaces, and adversarial verifiers test every change.

Claimed capabilities

Google makes the following claims about Carbon:

• Engineering: It performs well on tasks requiring multi-agent collaboration, including multi-package migrations, cross-subsystem root-cause investigations and complex architectural redesigns.
• Web development: It runs build-and-verify loops, recovers from errors, and checks both its code and the rendered UI.
• Visual work: Stronger visual design, spatial reasoning and attention to environmental detail carry over to 3D game development and creative coding.
• Enterprise work: For legal, finance and enterprise use, Google says Carbon brings higher-precision knowledge work, artifact generation and data analysis. It is said to follow strict schemas and formatting to produce spreadsheets, presentations and reports.

Benchmark

Google says existing benchmarks do not adequately capture how agents reason natively over dense files with charts, images, blueprints, schematics and tables. It partnered with Artificial Analysis to launch the Multimodal Knowledge Agents (MKA) benchmark. Google claims Carbon ranks #1 while using 3X fewer tokens than the next best model across 200 tasks testing multimodal performance in agentic workflows. No other benchmark scores were published. Context window, parameter count and training cutoff date are not yet disclosed.

Availability

• Developers: The Gemini API paid tier in AI Studio and Android Studio. Google AI Ultra subscribers can also use it in Google Antigravity and AI Studio.
• Enterprises: Gemini for business via the API or model drop-down, Google Antigravity, and specialized agents such as CodeMender and AlphaEvolve.
• Scientists and researchers: Applicants can seek access to both Argon and Carbon through Antigravity's highest usage tier, free for 6 months, onboarded on a rolling basis and subject to eligibility, capacity and selection.
• Cybersecurity defenders: Applicants can join the Fairwind Program for access to Carbon for cyber work on a rolling basis.

What this means

Google is running two tiers of Gemini 4 models in parallel rather than replacing one with the other. Argon stays the default recommendation for most software engineering and knowledge work, while Carbon is aimed at the hardest tasks. Developers choosing between them will need to weigh Carbon's per-token price against Argon's, though Argon's pricing is not given here.

The introductory rate runs only until November 30, so teams evaluating Carbon through the API should price their workloads against the $4/$20 rates that follow. The MKA result is Google's own claim on a benchmark it launched with Artificial Analysis, so the 3X token-efficiency figure is worth testing on real document-heavy agentic workloads. The /swarm preview shows how Google expects Carbon to be used: as the high-reasoning coordinator in multi-agent setups, with lighter models doing parallel work.