Google Opens Gemini 4 Carbon to All Paying API Developers at Argon-Level Pricing Until November 30

What happened

On November 10, Google made Gemini 4 Carbon, described as its strongest AI model to date, available to every developer with a paid Gemini API key, according to a post on dev.to. Before the wider release, the model had been limited to a small group of cybersecurity defenders through Google's Fairwind Program, after being announced on October 28.

Paid API users can now call the model from Google AI Studio and Android Studio, and business customers can reach it through Gemini for business.

Intro pricing ends November 30

The pricing is what makes the launch notable. Through November 30, Carbon costs $2 per million input tokens and $10 per million output tokens — identical to Gemini 4 Argon, the cheaper model Google released a few weeks earlier. On December 1, both figures double to $4 and $20, leaving roughly three weeks in which the flagship model and the standard one cost the same per token.

The dev.to write-up adds a caveat worth noting: Carbon is built to reason longer and run extra verification passes, so it consumes more tokens by design. Even at matching per-token rates, a job on Carbon can cost more than the same job on Argon.

Two models, two jobs

According to the post, Google positions the pair as complementary rather than competing. Argon remains the sensible default for routine coding and knowledge work, while Carbon is aimed at the demanding end of the spectrum: complex multi-step coding and debugging, extended agentic workflows, and open-ended research.

On performance, the post reports Google's claim that Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, and does so at less than half the cost per task of the next best outside model.

The standout example cited involves Signal: researchers used Carbon alongside other models to formally verify the code behind Signal's cryptographic protocol. Their agents reportedly generated millions of lines of Lean, a language for writing mathematical proofs, and established key security properties in weeks — work that has historically taken expert teams years.

Google is also previewing a feature called /swarm in Antigravity, its AI development platform. A single prompt spins up a coordinated team of agents that can be watched working in real time; Carbon takes the roles that need the deepest reasoning, such as architectural design and auditing, while faster and lighter models handle the remaining tasks in parallel.

Availability limits

Not everyone gets in. The dev.to post notes that Carbon is absent from the free tier of the Gemini API. On the consumer side, it is rolling out only to Google AI Ultra subscribers, appearing in the Gemini app, AI Mode in Search and Antigravity. Developers on lower plans have to go through the API to use it.

Why it matters

For anyone already paying for the Gemini API, the next three weeks amount to a low-cost test of a capability upgrade: Google's top model at everyday-model rates. That makes it a good moment to hand Carbon the jobs that keep getting postponed — messy multi-package migrations, bugs that cross several subsystems, refactors nobody wants to start.

The December increase matters more than it first appears. Any workflow that only makes financial sense at $2 and $10 per million tokens will stop making sense on December 1, so the practical advice from the post is to evaluate and budget at the doubled prices rather than the introductory ones. And because Carbon burns more tokens by design, the real per-task gap versus Argon will be wider than the headline rates suggest, even before the increase.

The launch also sketches Google's broader strategy: a two-tier lineup where an affordable default handles routine work and a pricier reasoning model is rented for the heavy jobs, with orchestration tools like /swarm stitching the tiers together. Whether that split pays off in practice is precisely what the discounted window lets developers find out.