Google has unveiled Gemini 4 Carbon, a new frontier AI model that it describes as its strongest yet for cybersecurity and long-running coding work. Unusually, the model is not opening up to everyone on day one. Google says Carbon will go first to a small group of cyber defenders, with access widening only after the company has gathered feedback and tightened its guardrails.

The model was announced on 28 October in a blog post by Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect at Google. It arrives just two weeks after Gemini 4 Argon. Reuters, in a report carried by The Hindu, framed the launch as Google stepping up its race with OpenAI and Anthropic to build more capable coding and security systems.

What Google is claiming

According to Google, Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59. The company is putting most of its emphasis on cyber defence, quoting benchmark results across the full vulnerability lifecycle:

• Finding flaws: 88.2% on SEC-bench Pro, which tests the discovery of real vulnerabilities in real-world software.
• Validating risk: 54.8% on ExploitGym, an exploitation benchmark that defenders use to decide which bugs to fix first.
• Fixing root causes: 71.0% on Collinear’s CWE-bench, which measures how well findings turn into working patches.
• Secure code: more than twice Argon’s success rate at generating secure code on internal benchmarks. Google says its Chrome team found Carbon more than 10 times better than another lab’s large frontier model at producing high-quality patches.

Google says the external benchmark results were measured without guardrails. Beyond security, it is pitching Carbon at “long-horizon” engineering. In its account, internal teams have watched the model keep working for days on multi-step tasks such as tracing subtle bugs through large codebases. Carbon also climbed 89 Elo points to fourth place on WebDev Arena.

Safety claims and a slower release

On safety, Google lists several figures. Successful cheating on Terminal-Bench 4.0 fell to 0.0%, with cheating attempts at 0.4%. High-severity misalignment alerts in live internal coding environments dropped by almost 80%. Unsupported claims were roughly halved compared with Argon, and the model recorded a 0.0% attack success rate on Gray Swan’s indirect prompt-injection benchmark. These are the company’s own numbers and have not yet been independently tested.

The rollout is the main story. Google says it is taking part in the US government’s voluntary process for testing models before release. Carbon is going to a subset of cyber defenders this week, and access will then expand in phases to trusted partners through Google’s Fairwind Program. Developers, businesses and consumers will get it later, through Google’s products, “as soon as possible”, with no date given. Pricing at launch matches Argon’s introductory rate of $2 per million input tokens and $10 per million output tokens. After the introductory period, that rises to $4 and $20.

Why it matters for India

The timing matters for India’s own AI policy debate. As Elets CIO has reported, the government is considering a separate safety framework for the most capable AI models under its proposed AI law. Pre-release evaluation, sandbox testing and human-control safeguards are among the options, and a consultation paper is expected soon. Google’s approach, which pairs voluntary government testing with gated early access for defenders, is an example of the kind of model policymakers in New Delhi are now weighing.

For Indian CIOs and CISOs, the practical point is that frontier AI is starting to arrive in tiers. The most capable cyber tools may go to vetted security teams well before they reach general enterprise API access. Security leaders running large application estates, or managing exposure across banking, telecom and public digital infrastructure, will want to know how such access is granted and whether Indian defenders are included in the early groups. Google has not said which organisations or countries are in the first cohort.

The other takeaway is how fast the field is moving. Two frontier releases from one lab in about a fortnight, each with big jumps in coding and security, puts pressure on enterprises to keep their governance and model-evaluation processes up to date. The same tools that help defenders find and patch flaws faster can make attackers faster too. That is why staged access, independent testing and clear national rules are likely to stay at the centre of India’s AI conversation in the months ahead.

Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter.

"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!