Google Releases Gemini 4 Carbon Following Argon’s Debut
Google has introduced its latest frontier artificial intelligence model, Gemini 4 Carbon, just two weeks after the release of Gemini 4 Argon. As reported by Živě.cz, Gemini 4 Argon recently reached paying users in Google AI Pro and AI Ultra tiers, and the firm has already moved to test its successor, Gemini 4 Carbon, with a limited group of cybersecurity experts. Gemini 4 Carbon is currently restricted to a select group of security researchers and cybersecurity professionals under the company’s Fairwind program. Google has not yet provided a specific timeline for when the model will be available to the general public or business customers. Koray Kavukcuoglu, senior vice president of Google DeepMind, who assumed leadership of the research laboratory following the departure of Demis Hassabis, described the new release as Google’s strongest model to date.
Performance Rankings in the Artificial Analysis Intelligence Index
The new model has quickly secured the top position in the Artificial Analysis Intelligence Index v5.0, achieving a score of 59 points. This ranking places Carbon at the top with 59 points, ahead of Anthropic’s Claude Opus 5.5 at 58 points and OpenAI’s GPT-6 Astra at 57 points, according to the Artificial Analysis Intelligence Index. The index, updated just one day before the model’s release, aggregates results from nine different tests. While Carbon leads the overall score, competitors maintain advantages in specific niches. In agentic programming, where the model operates directly within a tool, Claude Code with Opus 5.5 leads with 62 points, while Carbon using Google’s own Antigravity CLI follows with 59. Google’s internal benchmarking suggests Carbon achieves top results in 15 out of 17 tests. However, in the FrontierSWE 2 software engineering exam, GPT-6 Astra leads, and Claude Opus 5.5 remains the top model for controlling computer interfaces in the OSWorld-2.1 evaluation.
I did say we'd be releasing our models periodically…introducing Gemini 4 Carbon, our next frontier model. Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys… pic.twitter.com/GN9LqZ7Zcf
Technical Advancements in Coding and Security
Google has positioned Carbon as a specialist in long-horizon tasks—complex operations that require AI agents to work for hours or even days. The model shows significant improvements in identifying and patching code vulnerabilities. According to Vietnam.vn, Carbon achieved an 88.2% success rate in the SEC-bench Pro test for vulnerability discovery, 54.8% in ExploitGym for exploitation, and 71.0% in CWE-bench for repairs. In a test involving context handling up to one million tokens, the GraphWalks result was 96.2% for Carbon, compared to 71.8% for Astra and 66.8% for Opus. Internal testing by Google indicates that Carbon demonstrates higher reliability than its predecessor. The company reports that the model produces roughly half as many unsubstantiated claims as Argon. In tests conducted by Gray Swan regarding indirect prompt injection, Carbon successfully resisted all attempted attacks. Google claims that Carbon never successfully cheated in Terminal-Bench 4.0, attempting such shortcuts in only 0.4% of cases. The company also showcased the model’s versatility through a video demonstration of a humanoid robot learning to play cornhole using Gemini 4 Argon technology.
Pricing Structure for Future API Integration
Despite the performance gains, Google intends to maintain the same pricing model for Carbon as it did for Argon. When the model eventually moves into the API ecosystem, costs are set at $2 per million input tokens and $10 per million output tokens. A significant cost-saving feature remains for developers: input tokens stored in the cache will be 95% cheaper than the standard input rate. This pricing strategy positions Google’s latest offerings as notably more affordable than the current rates for competing models like GPT-6 Astra or Claude Opus, which have been priced significantly higher per million tokens in recent market reports.
Google Limits Gemini 4 Carbon Access for Safety Testing
Why is Gemini 4 Carbon currently unavailable to the public? Google is currently in a testing phase, limiting access to a small group of cybersecurity experts and partners under the Fairwind program to ensure the model’s safety and reliability before a wider rollout.
How does Carbon differ from the previously released Gemini 4 Argon? While both models share the same architecture and security framework, Carbon shows improved performance in technical benchmarks, generates fewer unsubstantiated claims, and is more resilient against adversarial prompt injection attacks.
When will regular users gain access to Gemini 4 Carbon? Google has not provided an official release date for the general public. Based on the rollout of Argon, which appeared for Pro and Ultra users approximately two weeks after its announcement, industry observers are tracking the timeline, but no firm date has been set for the public release of Carbon. Users are encouraged to sign in to their accounts to participate in ongoing engagement programs, such as holiday-themed prize draws offering up to 300,000 CZK.
More on Carbon
Introducing Gemini 4 Carbon, our latest frontier model. Carbon delivers improved performance across domains — ranking #1 on v5.0 of the @ArtificialAnlys Intelligence Index and sets a new state-of-the-art in cybersecurity and long-horizon capabilities. It’s rolling out starting… pic.twitter.com/oQrHzxvMSo
Announcing Gemini 4 Carbon, our strongest model with state-of-the-art cybersecurity and long horizon capabilities. Carbon represents a step up across most benchmarks, delivering strong frontier performance in autonomous coding and cyber defense. We’re first making Carbon… pic.twitter.com/jgrKXp3RYl
Related reading