Just a couple of weeks after Google released its Gemini 4 Argon frontier artificial intelligence (AI) model for consumers and enterprises, they have the successor ready. The Gemini 4 Carbon, the company says, is its strongest model yet, with state-of-the-art cybersecurity and long-horizon capabilities. After a sustained period of testing, as has become the norm with frontier AI models, Gemini 4 Carbon is now rolling out to subscribers.
In fact, Gemini 4 Carbon is now available across AI Studio, Gemini API, Antigravity, the Gemini app, and AI Mode in Search, but with a catch—you need to be a Google AI Ultra subscriber or via the Gemini API paid tier. In India, the Google AI Ultra plans are priced at ₹6,500 and ₹19,500 per month each, depending on usage and storage specifics.
“Gemini 4 Carbon is rolling out today. It’s built for sustained, long-horizon tasks, reasoning deeply through complex software engineering, cybersecurity, and scientific discovery,” said Sundar Pichai, chief executive officer of Google and Alphabet, in a post on X.
Earlier in the day, Anthropic also rolled out the Fable 5.2 and Mythos 5.2 models, which seem to be setting new goalposts in certain domains.
“Building on Gemini 4 Argon, Carbon illustrates the pace at which we’re pushing the frontier forward. Our researchers and models are working hand in hand to advance capabilities faster, while strengthening our safeguards in tandem,” said Koray Kavukcuoglu, senior vice president, Google DeepMind and chief AI architect, Google.
{{/usCountry}}“Building on Gemini 4 Argon, Carbon illustrates the pace at which we’re pushing the frontier forward. Our researchers and models are working hand in hand to advance capabilities faster, while strengthening our safeguards in tandem,” said Koray Kavukcuoglu, senior vice president, Google DeepMind and chief AI architect, Google.
{{/usCountry}}Gemini 4 Carbon was initially made available to a select group of cybersecurity experts. Google said at the time that it would gradually phase access to additional trusted partners through its Fairwind Program.
As part of the Fairwind Program, Google works with 650 partners as a cybersecurity defence, and fine-tunes model deployment and safety measures—these include Accenture, CrowdStrike, Snowflake and Wiz.
The frontier AI battles
When compared with Gemini 4 Argon, OpenAI’s GPT-6 Astra, as well as Anthropic’s Claude Opus 5.5 and Claude Fable 5.1 models, the Gemini 4 Argon leads across cybersecurity benchmarks including SEC-bench Pro (88.2% score, ahead of 85.4% for GPT-5 Astra), ExploitGYM (54.8%, and ahead of 42.4% for GPT-6 Astra) and CWE-bench (71% for Gemini 4 Carbon, ahead of 68% tied between Argon and GPT-6 Astra).
Mind you, the landscape has changed since Anthropic’s new frontier models rolled out today. They seem to be leading most benchmarks, Anthropic suggests.
The GraphWalks long-context benchmark also sees Gemini 4 Carbon score 96.2% and significantly ahead of 84.2% for the Gemini 4 Astra and 71.8% for GPT-6 Astra.
In the Artificial Intelligence Index 50, which is a collective of nice AI benchmark evaluations across tasks and categories, including agentic coding, frontier agentic capabilities, as well as measuring factual recall and hallucination across various economically relevant domains, Gemini 4 Carbon is now at the top of the leaderboard with a score of 59, ahead of Claude Opus 5.5 (score of 58) and GPT-6 Astra (57).
Some other prominent AI models released recently score significantly lower in this benchmark. Meta’s Muse Wonder Full scores 50, with the Kimi K3.1 Pro scoring 46 while Grok 4.7 scores 34.
“Designed to take on much harder, long-horizon engineering problems, Carbon is built for sustained agentic workflows. Internally, our teams have seen Carbon run for days to persist through dead ends to tackle complex, multi-step engineering tasks like root-causing and patching subtle bugs across large codebases and low-level systems,” said Kavukcuoglu.
Google insisted that Gemini 4 Carbon is building on the same multi-layered safety and security stack as the Argon, which includes combining model training, real-time misuse defences, prompt-injection resilience, and chain-of-thought monitoring.
They also claimed that in internal tests, Carbon reduced unsupported claims by approximately 2x over Argon, and saw an uptick in collaboration as it retrieved relevant sources and surfaced discrepancies to the user before acting.
In terms of cost, Gemini 4 Carbon will launch at a similar token usage introductory price as Argon—that is $2 per million input tokens and $10 per million output tokens. It is not clear how long this introductory pricing will last, but after that expiry, it will cost $4 per 1M input tokens and $20 per 1M output tokens.