Google made Gemini 4 Carbon, its most capable Gemini model, generally available on Nov. 10 to paying developers, enterprises and Google AI Ultra subscribers. The API launches at an introductory price of $2 per million input tokens and $10 per million output tokens, the same introductory rate as Gemini 4 Argon.
That price is temporary. After the introductory period ends on Nov. 30, 2026, Gemini 4 Carbon will cost $4 per million input tokens and $20 per million output tokens. Developers can reach it through the Gemini API paid tier in Google AI Studio and in Android Studio, while enterprises get it in Gemini for business, Google Antigravity and CodeMender. AI Ultra subscribers get Carbon in the Gemini app, AI Mode in Search and Antigravity.
Google is positioning Carbon above Argon rather than as a replacement. In the announcement, Koray Kavukcuoglu, SVP of Google DeepMind and Google's Chief AI Architect, said Argon remains the right default for most software engineering and knowledge work. Carbon is meant for the hardest jobs, such as complex multi-step coding, debugging and open-ended scientific research, and it spends more tokens than Argon exploring, reasoning and running verification loops.
Key facts
- 01Google made Gemini 4 Carbon available to paying developers, enterprises and AI Ultra subscribers on Nov. 10, 2026.
- 02Introductory API price is $2 per 1M input and $10 per 1M output tokens, rising to $4 and $20 after Nov. 30.
- 03Carbon ranked #1 on CursorBench 4.0, WANDR and AutomationBench at more than 2x lower cost per task than rivals.
- 04Per-attempt cyberattack breach rate fell to 0.8% on Carbon from 3.1% on Gemini 4 Argon.
- 05Carbon passed 92% of a 375-task agentic honesty benchmark, versus 70% for Argon.
The wider release follows a limited one. Two weeks earlier, Google gave Gemini 4 Carbon to cyber defenders through its Fairwind Program. The cyber-tuned version, Carbon for cyber, stays inside Fairwind, and security teams can apply for access on a rolling basis.
On Google's benchmark results, Carbon ranked first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, in each case at more than 2x lower cost per task than the next-best model from another company. It placed second on the new Artificial Analysis Multimodal Knowledge Agent benchmark at more than 6x lower average cost per task than the top model. In pre-release testing it ranked first of 23 frontier model configurations on Balyasny Asset Management's General Financial Intelligence evaluation and matched the best model on Hebbia's end-to-end financial agent test.
Cybersecurity is where Google claims the biggest lead. Accessed through Fairwind, Carbon leads on SEC-bench Pro for finding vulnerabilities, ExploitGym for exploiting them and CWE-bench v1 for patching them. In Project Fortify, Google's autonomous scanning and remediation system, Carbon hit 100% recall on scanning and 95% precision on triage on an internal benchmark, and improved on Argon's recall by more than 25% in open-source repository testing.
Google paired those capabilities with misuse numbers. In simulated attack environments, Carbon blocked 99% of attack trajectories from an adaptive red-teaming agent, and its per-attempt breach rate fell to 0.8% from 3.1% on Argon. Its success rate on Gray Swan's automated cyberattack queries was 2.17%, against 9.86% for Argon, and external tester Trajectory Labs found no new universal jailbreaks across 132 hours and more than 13,000 requests.
Carbon also scored higher on honesty. It passed 92% of a 375-task agentic honesty benchmark, compared with 70% for Argon, and declined to fill in missing experimental data 98% of the time. It reached 0% attack success on Gray Swan's indirect prompt injection benchmark, though Google said prompt injection robustness remains an area for improvement.
Scientists can apply for six months of free access to Argon and Carbon through Antigravity's highest usage tier. Google offered Argon access to 10,000 scientists four weeks ago. Early science results include Axiom Bio's finding that Carbon raised protein homolog retrieval 40-fold on average across seven low-homolog CASP17 families. Antigravity also gets a /swarm preview, in which Carbon handles architecture, coordination and auditing while lighter models run tasks in parallel.
Google said broader benchmarking of Gemini 4 Carbon on scientific work is still underway.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
