Google widened access to its Gemini 4 Carbon artificial intelligence model on Wednesday, opening the advanced system to developers, enterprise customers and high-tier consumers following an initial rollout to cybersecurity professionals two weeks ago.
The release marks the tech giant’s latest push into frontier computing, coming only a month after the debut of its predecessor, Gemini 4 Argon. While the company stated that Argon remains the preferred choice for standard software engineering and everyday knowledge tasks, Google described Carbon as its most capable architecture yet, specifically engineered for intricate multi-step coding, cybersecurity analysis and open-ended scientific exploration, The Verge reported.
Deploying Swarms of Digital Agents
Alongside the broader rollout, Google introduced a preview of an experimental “swarm” capability on Antigravity, its platform for agentic software creation. The functionality allows developers to transform an individual prompt into a synchronized network of AI agents working simultaneously on complex projects.
The company confirmed that Carbon is now available through Google AI Studio, Android Studio and Gemini for business. Paying subscribers to the Google AI Ultra plan—priced at $100 per month—can also access the tool through the Gemini application, Search’s AI Mode and the Antigravity developer environment, according to Engadget.
Google has also moved to court researchers by offering six months of complimentary access to 10,000 scientists. The company said thousands have already begun using the system, pointing to early breakthroughs in protein structure mapping that could accelerate pharmaceutical development, Engadget reported. During early cybersecurity tests in late October, participating defense firms deployed the model to identify software flaws and zero-day vulnerabilities in compressed timeframes.
Aggressive Pricing and Benchmark Claims
To spur rapid adoption, Google launched the model with introductory pricing of $2 per million input tokens and $10 per million output tokens, according to The Verge. Those rates sit at roughly one-fifth the price of OpenAI’s GPT-6 Astra and about one-fourth the cost of Anthropic’s Fable 5.2, which debuted Tuesday.
Even though Google plans to double its token rates when the introductory promotional period ends in late November, Carbon will remain significantly cheaper per token than competing frontier systems. However, overall expenses will depend on the volume of tokens each model consumes to finish a given assignment, The Verge noted.
Performance benchmarks presented by Google show Carbon challenging the industry’s top systems, although published metrics indicated it still trailed Anthropic’s Fable 5.2 on several key criteria, The Verge reported. Separately, CNET reported that Google claimed the model outperformed rival architectures across specialized benchmarks, including scientific reasoning, agentic coding and physics simulations for three-dimensional game engines.
Addressing growing industry concerns over defensive safeguards, the company reported that Carbon registered a zero percent success rate for prompt-injection attacks on the Gray Swan IPI benchmark test, according to CNET.
What Lies Ahead
The aggressive rollout schedule signals Google’s intent to challenge OpenAI and Anthropic directly at the highest tier of commercial computing after months in which competitors appeared to dominate frontier research.
Whether Google will eventually offer Gemini 4 Carbon to non-paying users or those enrolled in less expensive subscription tiers remains unclear. For now, the company has not revealed any timetable for a broader consumer distribution, leaving the model primarily in the hands of enterprise clients, researchers and premium tier subscribers.
