Google has unveiled its latest frontier model, Gemini 4 Carbon, and according to the company, it delivers "significant gains" in coding, agentic tasks, and cybersecurity. In a Wednesday announcement, Google touted Carbon's top spot on the Artificial Analysis Intelligence Index (v5.0) benchmark, where it edged out Claude Opus 5.5 and GPT-6 Astra. Because nothing says "we're winning" like a leaderboard.

The new model improves upon Argon, which Google released earlier this month - yes, that's right, earlier this month. The company is apparently moving at a pace that makes your phone's software update schedule look glacial. Carbon is currently only available to a select group of cybersecurity defense companies, but it will eventually expand to other partners in its Fairwind Program. So if you're not in the club yet, keep waiting.

Google claims Carbon represents the "pace at which we're pushing the frontier forward," and comes with a "significant leap" in cyberdefense. It's better at finding vulnerabilities and generating secure code, which is great news for anyone who enjoys their software not being held hostage by ransomware. Carbon currently holds the number one spot on SEC-bench Pro - a benchmark measuring an AI model's ability to discover real-world bugs - and ExploitGym, which measures how well a model can turn a vulnerability into exploits. Because nothing says "defense" like knowing exactly how to attack.

According to Google, Gemini 4 Carbon is built for completing longer tasks. Internal teams have reportedly seen it "run for days to persist through dead ends to tackle complex, multi-step engineering tasks." So it's essentially the AI equivalent of that one colleague who never gives up, even when everyone else has gone home. It's also better at avoiding shortcuts, resisting prompt injection attacks, and collaborating with users. And in a refreshing twist, Carbon offers more "transparency and honesty" by cutting "unsupported claims" by two times compared to Argon. Because an AI that admits it doesn't know something is apparently a feature now.

The model will launch at the same competitive introductory price as Argon: $2 per million input tokens and $10 per million output tokens. So your AI assistant can now find bugs, resist manipulation, and work for days on end - all for the low, low price of your monthly coffee budget. Google says it's a leap forward. We'll see if it lands.