Google opens Gemini 4 Carbon, its strongest model, to paying developers and AI Ultra subscribers

The introductory price doubles after November 30, and Anthropic's own table puts Fable 5.2 ahead on coding.

By Think Facility · November 11, 2026 · Google AI models

On November 10, 2026, Google made Gemini 4 Carbon, which it calls its strongest model, available to every paying developer and enterprise on the Gemini API, and started rolling it out to Google AI Ultra subscribers. The announcement came from Koray Kavukcuoglu, Google's chief AI architect.

Until then, only cyber defenders had it. Google announced Carbon on October 28 and gave it to a subset of them through its Fairwind Program, saying it wanted feedback and time to iterate on guardrails before a wider launch.

Model: Gemini 4 Carbon
Available to: paying developers and enterprises, and Google AI Ultra subscribers
Introductory price: $2 per million input tokens and $10 per million output tokens
After November 30, 2026: $4 per 1M input tokens and $20 per 1M output tokens
New in Antigravity: a preview of /swarm

Who can use it, and where

Developers get Carbon in Google AI Studio and Gemini for business, at the same introductory price as Gemini 4 Argon. Google says Argon is still the right pick for most software and knowledge work. Carbon is for harder jobs, and it spends more tokens than Argon exploring and checking its work.

A large canopy roof of white tiles on a red and white steel frame, wrapped in scaffolding, behind a field of dry grass under a blue sky — A Google building under construction at the Googleplex in Mountain View, California, June 2019. Photo: Gregory Varnum, CC BY-SA 4.0, via Wikimedia Commons

For everyone else, it's an Ultra feature for now. Ultra subscribers get Carbon in the Gemini app, in AI Mode in Search and in Google Antigravity.

Antigravity also gets a preview of /swarm, which turns one prompt into a team of agents you can watch as they work. Google suggests giving Carbon the roles that need the deepest reasoning (design, coordination, auditing) and pairing it with faster, lighter models working in parallel.

a multi-agent Carbon team using Google Antigravity’s /swarm feature built an interactive 906K-triangle Three.js reconstruction of Google’s Platform 37 office across 10 overnight loops

From Gemini 4 Carbon is now available

How it compares with Fable 5.2

Google's post doesn't put Carbon next to any named rival. It says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, at under half the cost per task of the next best outside model.

Anthropic's Fable 5.2 launch page does include a Carbon column. Here are the three tests where both models have a score.

FrontierCode 2.0: Fable 5.2 49.1%, Gemini 4 Carbon 41.1% (Medium effort)
FrontierSWE v2: Fable 5.2 69.9%, Gemini 4 Carbon 61.2%
AutomationBench: Fable 5.2 51.4%, Gemini 4 Carbon 53.9%

On Anthropic's numbers, Fable 5.2 leads on both coding tests and Carbon leads on AutomationBench, which fits Google's own claim there. Anthropic's FrontierCode score for Carbon is at medium effort, so it may not be Carbon's best. These are one lab's figures for a rival's model, and I'd want independent runs before reading much into a gap of a few points.

The testing before the wide release

Google says it took part in the U.S. government's voluntary process for pre-release model testing. In simulated cyberattack environments, it says, Carbon's safeguards blocked 99% of attack paths from an adaptive red-teaming agent and cut the per-attempt breach rate to 0.8%, down from 3.1% on Argon.

Outside testers had a go too. Trajectory Labs found zero new universal jailbreaks across 132 hours and over 13,000 requests. On Google's 375-task agentic honesty benchmark, Carbon passed 92% of tasks, against 70% for Argon, and declined to fill in missing experimental data 98% of the time.

What it means for scientists and security teams

The widest new access outside the tech industry is for researchers. Google's program giving 10,000 scientists free use of Argon in Antigravity for six months now includes Carbon, and applications are still open, subject to eligibility and capacity.

Google gives two examples of what that work looks like. Axiom Bio, a drug toxicity research company, found mechanisms of toxicity with Carbon that it hadn't found before, Google says, and is moving its production science work to Carbon and Argon. In protein research, Carbon raised the number of related sequences found across 7 hard protein families by 40-fold on average, which gives structure prediction models like AlphaFold 3 more to work with.

Security teams get a mixed picture. Wiz's Scan for Good program used Carbon on two major SaaS providers and found 0-day vulnerabilities in under two hours, according to Google, and the affected companies fixed them. The version with Carbon's strongest cyber skills stays behind the Fairwind Program, which defenders still have to apply to.

The introductory price ends on November 30, 2026, when Carbon moves to $4 per million input tokens and $20 per million output tokens.