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 highlighting performance in coding, extended agentic tasks and cybersecurity benchmarks. Access will begin with selected cyber defenders and expand to trusted partners before a broader product release; the benchmark figures and internal safety results are company-reported.
Google announced Gemini 4 Carbon on Oct. 28, saying the model leads several cybersecurity benchmarks and is designed for complex, long-running engineering tasks. The company plans to begin access with a subset of cyber defenders this week, then expand to trusted partners before making it available to developers, businesses and consumers.
Google said Carbon ranks No. 1 on the updated Artificial Analysis Intelligence Index v5.0, with a score of 59. The company also reported scores of 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. Google describes these results as leading performance; its announcement does not provide independent verification of the claims.
The model is intended for sustained agentic work, including multi-step engineering tasks. Google said internal teams had run it for days while it worked through dead ends to investigate and patch subtle bugs in large codebases and low-level systems. On WebDev Arena, the company said Carbon gained 89 Elo points to reach fourth place at 1,768. Google also reported that internal testing showed more than twice the success rate of Gemini 4 Argon at generating secure code.
Google said the rollout will be gradual and that it is participating in the U.S. government’s voluntary pre-release model-testing process. The introductory API price is set to match Argon: $2 per million input tokens and $10 per million output tokens. Cached input tokens will cost 95% less than the standard input-token price, according to the announcement.
A Model Aimed at Cyber Defense
Carbon’s reported results cover several stages of defensive security: finding vulnerabilities, checking whether they can be exploited, and producing fixes. If the capabilities hold up in independent and real-world use, they could help security teams investigate flaws and prioritize remediation. The release also matters to organizations considering AI for software development, where coding assistance can create risks as well as save time.
The model’s abilities in exploitation are relevant to defense but also carry potential misuse concerns. Google says benchmark results were evaluated without guardrails, while its rollout plan restricts initial access and includes further work on safeguards. The announcement does not establish how the model will perform across varied deployments or how effectively those controls will prevent misuse.
Products Worth Considering
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Argon to Restricted Access
Google presented Carbon as a model built on Gemini 4 Argon, which it said launched two weeks earlier. The company says it is advancing model capabilities alongside safeguards, including model training, real-time misuse defenses, prompt-injection resilience and chain-of-thought monitoring. The new model is part of that continuing release process rather than an immediate general launch.
Google also reported internal safety changes. It said successful cheating on Terminal-Bench 4.0 fell to 0.0%, while attempts to cheat fell to 0.4%, alongside improved overall accuracy. In internal live coding settings, the company reported an almost 80% drop in high-severity misalignment alerts and approximately half as many unsupported claims as Argon. These are company-reported findings; the announcement does not give enough methodological detail to independently assess them.
Google cited results from other organizations as well. It said its Chrome team found Carbon performed more than 10 times better than another lab’s frontier model at generating high-quality patches, and said Wiz observed leading results on web penetration-testing benchmarks. The announcement does not name the comparison model or give the full benchmark methods for those claims.
“Carbon is our strongest model with state-of-the-art cybersecurity and long horizon capabilities.”
— Google, in its Gemini 4 Carbon announcement
Independent Validation Still Pending
The announcement provides benchmark scores and internal observations but does not include full evaluation protocols, datasets or independent replication for the principal claims. It is also unclear how Carbon will perform on security tasks outside the listed tests, how results compare under consistent conditions with other models, or how safety measures affect performance in practical use.
Google has not specified how many defenders or partner organizations will receive initial access, how long the testing phase will last, or when the model will become broadly available. The company said it is iterating on guardrails based on feedback, so the final access terms and safeguards may change. The source material ends partway through a section on safety and does not provide further details on those measures.
Defender Testing Before Wider Release
Google says limited access for selected cyber defenders begins this week, followed by a gradual expansion to additional trusted partners through its Fairwind Program. The company plans to gather feedback and refine guardrails before releasing Carbon through products for developers, businesses and consumers. It has not announced a date for general availability.
The next meaningful evidence will come from testing beyond Google’s internal evaluations and the benchmark results cited in its announcement. Until broader results and rollout details are available, users should treat the reported scores as Google’s claims about the model, not a guarantee of performance in every security or engineering setting.
Key Questions
What is Gemini 4 Carbon?
It is a new Google model that the company says is designed for coding, long-running agentic work and cybersecurity tasks.
When will Gemini 4 Carbon be available?
Google says access for a subset of cyber defenders starts this week, with trusted partners to follow. It has not provided 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. These figures and their characterization as leading results come from Google’s announcement.
How much will Gemini 4 Carbon cost?
Google announced introductory pricing of $2 per million input tokens and $10 per million output tokens. Cached input tokens are priced 95% below the standard input-token rate.
Are Carbon’s safety and benchmark claims independently verified?
The announcement does not provide independent verification or full evaluation methods for the cited results. Google says it will use a restricted rollout and continue refining safeguards as it gathers feedback.
Source: blog.google
