Google opens Gemini 4 Carbon to all paying developers at Argon pricing

The latest:

Gemini 4 Carbon is now available to every paying developer through the Gemini API in Google AI Studio and Android Studio, Google said on November 10. The company priced it at $2 per million input tokens and $10 per million output tokens until November 30, 2026 — identical to Gemini 4 Argon — before rates double. Google AI Ultra subscribers get it in the Gemini app and AI Mode in Search.

Details:

• Where it runs: Google said Carbon is live across Gemini for business, through both the API and the model picker, as well as in Google Antigravity and CodeMender. Ultra subscribers can test it in Antigravity and AI Studio. The announcement was written by Koray Kavukcuoglu, senior vice president at Google DeepMind and chief AI architect.
• The price cliff: After November 30, 2026, pricing rises to $4 per million input tokens and $20 per million output tokens — double the introductory rate. Google framed the opening period as matching Argon’s cost exactly, giving teams a roughly three-week window to benchmark the stronger model at the cheaper tier.
• Benchmark claims: According to Google, Carbon ranks first on CursorBench 4.0, Perplexity’s WANDR and Zapier’s AutomationBench, at more than twice lower cost per task than the next best external model. It placed second on the Multimodal Knowledge Agent evaluation with Artificial Analysis, at more than 6 times lower average cost per task than the leader.
• Finance and science: Google said Carbon matched the best model tested on Hebbia’s pre-release E2E Financial Agentic evaluation and topped 23 frontier model configurations on Balyasny’s General Financial Intelligence benchmark. It cited Axiom Bio calling Carbon the leading model in translational science, with roughly 40 times more protein homologs retrieved across 7 families in the low-homology CASP17 set.
• Agent swarms: A preview feature turns a single request into a coordinated agent team through the /swarm command in Google Antigravity, trackable in real time. Google said Carbon handles architecture, coordination and review while faster models work in parallel isolated workspaces, with adversarial reviewers testing every change.
• What swarms built: Google’s examples include an interactive Eiffel Tower gallery built in Three.js from one prompt, and a reconstruction of the Platform 37 office at 906,000 triangles, generated from 7 PDF files containing 206 drawings across 10 overnight cycles.
• Security defenders: Carbon reached cybersecurity defenders two weeks before launch through the Fairwind program, with access continuing in batches. Google said it leads SEC-bench Pro for discovery, ExploitGym for exploitation and CWE-bench v1 for patching, and that internal Project Fortify results showed 100% scanning recall and 95% triage precision.
• Live findings: In Wiz Scan for Good, Google said Carbon found zero-day vulnerabilities at two major SaaS providers in under two hours. The Google Cloud Vulnerability Research team used it to confirm exploitability of two flaws in open-source software; Google withheld details pending responsible disclosure.
• Safety figures: Google reported Carbon blocked 99% of attack paths from an adaptive red-team agent, with a 0.8% per-attempt breach rate against 3.1% for Argon. External testing by Trajectory Labs found no new universal jailbreaks across 132 hours and more than 13,000 requests. On indirect prompt injection, Google reported a 0% attack success rate while noting room to improve.
• Scientist access: Google opened applications for 6 months of free access to Argon and Carbon at the top Antigravity usage tier, granted in batches subject to eligibility, capacity and selection. Applications had opened 4 weeks earlier for 10,000 scientists to access Argon for free.

Between the lines:

Pricing Carbon at Argon’s rate for three weeks lowers the switching cost for teams already building on the cheaper tier, and the per-task comparisons Google published on CursorBench, WANDR and AutomationBench are framed around cost, not capability alone. The safety gap against Argon — a 0.8% versus 3.1% breach rate — positions Carbon as the upgrade path rather than a parallel option.

What’s next

November 30, 2026 is the hard marker: introductory pricing ends and rates double. Also watch the staged Fairwind rollout to defenders, the batched scientist access grants, and disclosure of the two open-source flaws Google is holding back.