Gemini 4 Carbon: Google New No. 1 AI, Nobody Can Use Yet

What Carbon is: strongest model yet, built for long engineering and security work

Carbon "is our strongest model with state-of-the-art cybersecurity and long horizon capabilities," Google DeepMind's Koray Kavukcuoglu wrote in the announcement, according to Skimwire. Google says it ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59, and leads three security benchmarks: 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploiting them, and 71.0% on CWE-bench for patching them, according to Skimwire. Google notes those security scores were run without guardrails, according to Skimwire.

The model is built for completing longer engineering tasks and based on internal tests, it can run for days to persist through dead ends in order to carry out complex, multi-step actions, according to Indian Express. It is also better at avoiding shortcuts, resisting prompt injection attacks, vulnerability detection, validation, remediation, and collaborating with users, according to Google, via Indian Express. "Internally, our teams have seen Carbon run for days," Google wrote in the announcement, authored by Koray Kavukcuoglu, Google DeepMind SVP and Google's chief AI architect, according to VentureBeat.

Google has unveiled Gemini 4 Carbon, its latest artificial intelligence (AI) model, which ranks first on the Artificial Analysis Intelligence Index with a score of 59 and shows improved capabilities in detecting software vulnerabilities, exploiting them for testing and developing fixes, according to ET CISO. According to a blog post, the model will initially be made available to a select group of cybersecurity defenders this week, before Google gradually expands access to trusted partners through its Fairwind Program, according to ET CISO. The company plans to make it available to developers, businesses and consumers after gathering feedback and improving safeguards, according to ET CISO.

Carbon follows Gemini 4 Argon, launched two weeks ago, and is designed to handle complex, multi-step engineering tasks over longer periods, according to ET CISO. Google said Carbon achieved more than twice the success rate of Argon in generating secure code on internal benchmarks, according to ET CISO. Google said Carbon reduced its rate of successful cheating on Terminal-Bench 4.0 to zero, with attempts to cheat falling to 0.4 per cent, according to ET CISO. "The Chrome team has reported that Carbon performed more than 10X better than another lab's large frontier model when generating high-quality patches," Koray Kavukcuoglu wrote, according to SiliconANGLE. Argon, the company's previous flagship LLM, includes guardrails designed to mitigate hallucinations and cyberattacks, according to SiliconANGLE. Google implemented an improved version of that feature set in Carbon, according to SiliconANGLE.

Carbon is more transparent as well, cutting unsupported claims by two times when compared to Argon, according to Indian Express. Carbon improves substantially over Argon in coding and knowledge work, but trails several rivals on professional tasks and records a higher hallucination rate than Argon on a factual-knowledge test, according to VentureBeat — so Carbon is not better at everything; it's better at long engineering and security, worse at some professional tasks and factual knowledge, per VentureBeat's reading of Google's own numbers.

Benchmarks: #1 on Intelligence Index v5.0, top on security, but mostly Google's numbers

Independent evaluator Artificial Analysis places Carbon, using high reasoning, first on its Intelligence Index v5.0 with 59 points, ahead of Claude Opus 5.5 at 58, GPT-6 Astra at 57 and Argon at 50, according to VentureBeat. Separately, Carbon posts the highest result in 15 of the 17 benchmark rows in Google's announcement, although OpenAI retains an advantage on one software engineering test and Anthropic leads on one computer-use evaluation, according to VentureBeat. Google tested Carbon on a new version of the index that was publicly released today, according to SiliconANGLE. It includes two additional benchmarks and removes a number of saturated questions, questions deemed too easy for frontier models, according to SiliconANGLE.

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 Intelligence Index at 59, according to Sundar Pichai on X, via Indian Express. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching, according to Pichai, via Indian Express. As we did with Argon, we're going to roll it out gradually + responsibly, according to Pichai, via Indian Express. Some trusted cyber defenders will get access starting this week, according to Pichai, via Indian Express. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight! according to Pichai, via Indian Express.

Carbon secured the top stop on the Artificial Analysis Intelligence Index (v5.0) benchmark, with Claude Opus 5.5 and GPT-6 Astra trailing behind, according to Indian Express. Carbon also leads on SEC-bench Pro, a cybersecurity benchmark that measures an AI model's ability to discover real-world bugs, and ExploitGym, which measures exploitation, according to Indian Express. The new AI model achieves 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71% on CWE-bench, with access initially limited to cyber defenders and trusted partners, according to MoneyControl. Google has announced Gemini 4 Carbon, its latest frontier AI model designed for coding, complex engineering tasks and cybersecurity, according to MoneyControl. The model builds on Gemini 4 Argon, which launched two weeks ago, and introduces improvements in vulnerability detection, exploitation, remediation and long-running agentic workflows, according to MoneyControl.

There is an important condition attached: Google says the external cybersecurity evaluations were conducted "without guardrails," according to VentureBeat. The announcement therefore does not establish that customers will obtain the same scores under whatever restrictions accompany their deployment, according to VentureBeat. Google didn't release a technical report yet, nor say if Gemini 4 Argon offers a new model architecture, according to VentureBeat. Nor did it disclose a parameter count, according to VentureBeat. So treat Carbon's security scores as lab scores without guardrails, not as guaranteed scores in your deployment with guardrails on — and note that Google hasn't published a technical report or parameter count yet.

Access and price: Fairwind defenders first, $2/$10 introductory, $4/$20 later

In terms of its release, Google said that it has taken a "deliberative approach" by making Gemini 4 Carbon currently only available to a select group of cybersecurity defense companies, but it will eventually expand to other partners in its Fairwind Program as well, according to Indian Express. At launch, Carbon will be available at the same competitive introductory price as Argon, that is, $2 per million input tokens and $10 per million output tokens, according to Indian Express. Carbon's debut was accompanied by Google's latest earnings report which showed stellar third-quarter growth in its cloud computing business but particularly weak growth in Google's search advertising revenues, according to Indian Express.

Google will initially make Gemini 4 Carbon available to a subset of cyber defenders this week, according to MoneyControl. Access will gradually expand to additional trusted partners through its Fairwind Program, according to MoneyControl. The company said it is also participating in the US government's voluntary process for pre-release model testing, according to MoneyControl. Google says it's taking part in the U.S. government's voluntary pre-release testing and will widen access gradually through Fairwind before opening Carbon to developers, businesses and consumers "as soon as possible," according to Skimwire. The company will make Carbon available later this week through its Fairwind Program, which gives organizations such as internet providers early access to its LLMs, according to SiliconANGLE. Google plans to refine Carbon's guardrails based on feedback from early adopters before making it broadly available, according to SiliconANGLE.

Google says Carbon will launch at $2 per million input tokens and $10 per million output tokens, with a 95% discount for cached inputs, according to VentureBeat. Those rates are temporary: a footnote specifies subsequent prices of $4 and $20, respectively, according to VentureBeat. Google hasn't specified in its release blog post when Carbon's introductory period will end, according to VentureBeat. Carbon will launch at Argon's introductory rate of $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 when the introductory period ends, according to Skimwire. Google hasn't said when that is, according to Skimwire. The model will be offered at the same introductory price as Argon: USD 2 per million input tokens and USD 10 per million output tokens, with cached input tokens priced at a 95 per cent discount, according to ET CISO. Google said it would expand access gradually while continuing to test the model and strengthen safeguards before wider release, according to ET CISO.

So the price story is the same as Argon (per our Article 35): $2/$10 through the introductory period, $4/$20 after, cached input 95% off — but for Carbon, Google hasn't said when the introductory period ends, unlike Argon where the doubling was dated 1 December 2026. Budget Carbon at $4/$20 if you're planning beyond the introductory window, because the introductory rate is temporary and undated.

What to use instead, and verdict: Argon now, Carbon later, don't wait

Argon, which Google opened more broadly on October 13, remains the model customers can evaluate through its existing public offering, according to VentureBeat. So if you need a Gemini 4 model today, use Argon — it's broadly available, $2/$10 through 30 November 2026, $4/$20 from 1 December, 1M unified context, 1M output with long decode, per Article 35. If you need long-horizon security work — finding, exploiting, patching vulnerabilities over days, not hours — Carbon is the better model on paper, but you can't use it yet unless you're a Fairwind cyber defender. If you're a cyber defender, apply through Fairwind; if you're a developer, business, or consumer, wait for "as soon as possible" — no date given — and use Argon, GPT-6.1 Terra ($1/$5, 1M context, per Article 3), or Claude Opus 5 ($5/$25, per Article 21) in the meantime, depending on your task. Carbon's #1 score is real on Intelligence Index v5.0 (59 vs Opus 5.5's 58, Astra's 57, Argon's 50), but its security scores were without guardrails, no technical report or parameter count yet, and it trails on some professional tasks and factual knowledge — so it's a specialist, not a universal winner. When Carbon opens broadly, budget at $4/$20, test it on your own security tasks with guardrails on, and compare it to Argon on your workload before you switch.