TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Google announced Gemini 4 Carbon on Oct. 28, 2026, describing it as its strongest model and reporting leading results on benchmarks for vulnerability discovery, exploitation and patching. Access will begin with selected cyber defenders and trusted partners; a general release date has not been given.
Google announced Gemini 4 Carbon on Oct. 28, saying the model leads several cybersecurity benchmarks and is designed for long-horizon engineering work. The company plans to begin access this week with a subset of cyber defenders, then expand to trusted partners before making the model available to developers, businesses and consumers.
Google reported scores of 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for validating whether flaws can be exploited, and 71.0% on CWE-bench for producing fixes. The company said the evaluations were run without guardrails. These are benchmark results reported by Google; the announcement does not provide independent verification or full evaluation details.
The company also said Carbon ranked No. 1 on the updated Artificial Analysis Intelligence Index v5.0, with a score of 59, as detailed in the AI ranking results. On WebDev Arena, Google reported a score of 1,768 and fourth place, an increase of 89 Elo points. It said internal testing showed more than twice the success rate of Gemini 4 Argon at generating secure code, while Chrome’s team reported Carbon produced high-quality patches more than 10 times better than an unnamed model from another lab. Those comparisons are company-reported and their methods are not detailed in the announcement.
Google describes Carbon as built for sustained agentic workflows, including work that may involve tracing and fixing subtle bugs across large codebases. The company said internal teams had seen the model continue working for days on difficult, multi-step engineering tasks. Carbon will use the same introductory API pricing as Argon: $2 per million input tokens and $10 per million output tokens, with cached input tokens priced 95% below the input rate.
A Restricted Start for Cyber Defense
Carbon’s reported results span several stages of defensive security: locating software weaknesses, checking whether a flaw is exploitable, and generating a patch. If those capabilities perform reliably outside Google’s evaluations, they could help security teams investigate and remediate vulnerabilities more quickly. The initial rollout to defenders also gives Google a way to gather feedback on a model whose capabilities can be relevant to both defense and misuse.
The claims remain tied to Google’s announcement rather than an independent assessment. Benchmark performance does not by itself establish how well a model works on a company’s own systems, how often its findings are useful, or whether generated patches introduce new problems. The staged release and safety testing will matter as much as the headline scores.
Products Worth Considering
DJI Osmo Pocket 3 Vlogging Camera with 1-Inch CMOS Sensor and 4K Video
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Apple iPad Air (4th Generation, 10.9-inch, Wi-Fi, 256GB) – Space Gray
As an affiliate, we earn on qualifying purchases.
Built After Gemini 4 Argon
Carbon follows Gemini 4 Argon, which Google said launched two weeks earlier. The company presents Carbon as a further step in the Gemini 4 series, with work focused on coding, agentic and scientific tasks, full-stack web development and multimodal understanding. Google says it is advancing model capabilities alongside safeguards.
The announcement describes a safety stack combining model training, real-time misuse defenses, prompt-injection resilience and chain-of-thought monitoring. Google also reported internal alignment indicators, including an almost 80% drop in high-severity misalignment alerts in live coding environments and a roughly twofold reduction in unsupported claims compared with Argon. These are internal findings, not results from a published external trial. Google said it is participating in the U.S. government’s voluntary process for pre-release model testing.
“Gemini 4 Carbon is our strongest model with state-of-the-art cybersecurity and long horizon capabilities.”
— Google, in its Oct. 28 announcement
Independent Testing and Access Details
Google has not provided a date for general availability or a schedule for each stage of the Fairwind Program rollout. The announcement also does not specify how many defenders or partners will receive access, what eligibility criteria apply, or which product surfaces will offer Carbon later.
It remains unclear how the reported benchmark scores compare under independent testing and how they translate to operational results in different environments. Google says several external benchmark evaluations were conducted without guardrails, but the announcement does not explain how safeguards may affect performance in deployed use. The company’s internal figures on alignment, unsupported claims and coding behavior also have not been independently assessed in the material provided.
Defender Testing Before Wider Release
Google says it will start providing Carbon to a subset of cyber defenders this week and gradually extend access to additional trusted partners through the Fairwind Program. The company plans to collect feedback and refine guardrails before offering the model to developers, businesses and consumers through its products, but has not set a public timetable.
Further information to watch for includes rollout eligibility, published safety and evaluation details, and evidence from partners using Carbon in practical defensive work. Until those details emerge, the benchmark scores and internal comparisons should be treated as claims reported by Google rather than independent findings.
Key Questions
What is Gemini 4 Carbon?
Gemini 4 Carbon is a model Google announced on Oct. 28, 2026. The company says it is designed for coding, long-running engineering tasks and cybersecurity work.
When will people be able to use Carbon?
Google says access for a subset of cyber defenders begins this week, followed by a gradual rollout to trusted partners. It has not announced a date for general availability.
What cybersecurity results did Google report?
Google reported scores of 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench. The company says these measure vulnerability discovery, exploit validation and patching, respectively; the announcement does not provide independent verification.
How much will Gemini 4 Carbon cost?
Google announced introductory API prices of $2 per million input tokens and $10 per million output tokens, matching Argon. Cached input tokens will cost 95% less than the input-token rate.
Are Carbon’s safety claims independently verified?
The cited figures on secure code generation, misalignment alerts and unsupported claims are Google-reported internal results. The announcement does not provide an independent assessment of those findings.
Source: blog.google
