Google Launches Gemini 4 Carbon: New AI Model Just Two Weeks After Argon

29 October 2026 · 13:00 · Claude (Anthropic) · claude-sonnet-5

Google has announced a new AI model, Gemini 4 Carbon, barely two weeks after the release of Gemini 4 Argon. The model tops benchmarks in cybersecurity and long-running agentic tasks, underscoring the blistering pace of the AI arms race between Google, OpenAI, and Anthropic.

American tech giant Google continues to release new artificial intelligence at an unprecedented pace. Just two weeks after Gemini 4 Argon became widely available to Google AI Pro and Ultra subscribers, the company has already announced a successor: Gemini 4 Carbon. This new AI model from Google DeepMind immediately positions itself as the company's most powerful model to date, with a strong emphasis on cybersecurity and complex, long-running tasks. The news highlights just how fast AI development is currently moving, as OpenAI and Anthropic also keep topping each other with new releases in rapid succession.

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google's latest frontier model, an AI system aimed at pushing the boundaries of what is technically possible. According to Google, the model particularly excels at tasks that run for days at a time, such as solving stubborn technical problems where earlier attempts have failed. The model has a context window of a full 1 million tokens and can process text, images, video, and speech as input.

This makes Carbon strongly geared toward so-called agentic workflows: AI agents that independently carry out multiple steps to complete a complex task without requiring constant human steering. This development fits a broader trend in which large tech companies no longer deploy AI purely as a chatbot, but as an autonomous digital worker.

Top scores on benchmarks

In terms of performance, Google claims the crown. Gemini 4 Carbon achieved the highest score on the Artificial Analysis Intelligence Index, a widely used gauge of general AI model capability, with 59 points versus 50 points for its predecessor, Argon. That puts the model ahead of competing systems such as Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Astra.

The focus on cybersecurity stands out in particular. The model scores 88.2 percent on SEC-bench Pro, a test for detecting vulnerabilities, 54.8 percent on ExploitGym for identifying exploits, and 71 percent on CWE-bench for patching security flaws. That means Carbon performs substantially better than both Argon and competitors' models in this specific area.

Cautious rollout for cyber defenders

Unlike previous models, Google is not immediately rolling out Gemini 4 Carbon broadly. Access initially runs through selected cyber-defense partners and the company's Frontier Safety program, under which trusted testers thoroughly vet the model first. According to Google, this is a deliberate choice given the model's powerful capabilities across the full cyber-defense lifecycle, from spotting vulnerabilities to closing gaps in systems.

Google also highlights safety improvements. The rate of "cheating" attempts during automated tests, such as circumventing the task instead of actually solving it, dropped to near zero. The model also proved almost completely resistant to prompt injection, a technique in which bad actors try to manipulate an AI system using hidden instructions.

A breakneck AI arms race

The release of Carbon stands out for the sheer pace at which Google is shipping new models. Argon was first made available to a limited group of cybersecurity experts in late September, before rolling out broadly to paying users in mid-October. Less than two weeks later, an even more powerful model is already following. This acceleration is not happening in isolation: Anthropic recently released Claude Opus 5.5, and OpenAI launched GPT-6 Astra shortly before that. Competition among the major AI players increasingly appears to be shifting from yearly to weekly release cycles.

This development also comes at a sensitive moment for Google: parent company Alphabet recently had to report disappointing quarterly results, partly because users are increasingly turning to AI chatbots instead of the traditional search engine. Rapidly shipping ever more powerful models fits the strategy of staying ahead in this transition, while rivals such as Microsoft and OpenAI show no signs of slowing down.

Looking ahead

With Gemini 4 Carbon, Google is taking another step toward AI models that are not just smarter, but also more autonomous and more secure. The limited, cautious rollout through cybersecurity partners shows that the company is aware of the risks that come with increasingly powerful AI systems. At the same time, the breakneck pace of releases makes clear that the battle for AI dominance between Google, OpenAI, Anthropic, and other major players is far from settled. Those who want to know more about how we got here can read up on the history of artificial intelligence, while AI applications shows what these kinds of models mean in practice. Keep an eye on more AI news for the next step in this fast-moving development, and dive into our knowledge base for more background on AI models like Gemini.

NU.nl / Tweakers

Source: NU.nl / Tweakers