Quick Answer
Gemini 4 Carbon is Google’s new cybersecurity focused AI model, announced October 28, 2026. Initial access is limited to select defenders and Fairwind partners, while broader availability has no public date. Google lists introductory API rates of $2 per million input tokens and $10 per million output tokens. Organizations should treat Google’s performance and safety figures as self reported benchmarks rather than independent results.
Key Takeaways
- Google announced Gemini 4 Carbon on October 28, 2026, as a model for cybersecurity and long horizon engineering work.
- Gemini 4 Carbon initially reaches a subset of cybersecurity defenders and trusted Fairwind Program partners.
- Google reports 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench.
- Introductory API pricing is $2 per million input tokens and $10 per million output tokens.
- Google has not published a public date for consumer, developer, or business availability.
What is Gemini 4 Carbon?
Gemini 4 Carbon is Google’s newly announced AI model for cybersecurity work and long horizon engineering tasks. Google introduced the model on October 28, 2026, and describes Gemini 4 Carbon as its “strongest model,” with an emphasis on finding vulnerabilities, supporting exploitation work, patching code, and generating more secure software. Google’s Gemini 4 Carbon announcement sets out the company’s intended use cases and initial access plan.
Gemini 4 Carbon is part of the Gemini 4 generation, rather than an update to the earlier Gemini 4 Argon release. That distinction matters because Google positions Carbon around security and engineering workflows, while Gemini 4 Argon has a separate rollout path for paying subscribers. TechJournal previously covered Gemini 4 Argon’s subscriber rollout, which remains a different product availability decision from Carbon’s limited initial release.
Google’s description does not mean Gemini 4 Carbon can replace a security team or safely make production changes without review. AI generated code and vulnerability findings require validation because an incorrect finding can waste investigation time, while an incorrect patch can create outages or introduce a new security weakness. The practical use case is supervised assistance for trained defenders and engineers.
Who can access Gemini 4 Carbon now?
Gemini 4 Carbon is initially available only to a subset of cybersecurity defenders, with access planned to expand to trusted partners in Google’s Fairwind Program. Google DeepMind’s Gemini model page directs cybersecurity defenders to apply for Fairwind, which is the clearest current route for organizations seeking access. Google DeepMind’s Gemini model page identifies Gemini 4 Carbon as part of the Gemini 4 family and points eligible defenders toward the program.
Gemini 4 Carbon does not yet have a published public launch date for ordinary consumer accounts, general developers, or all business customers. Google says those groups will receive the model through its product surfaces “as soon as possible,” but the announcement does not define a schedule, region list, plan requirement, or product by product availability timeline.
Organizations should avoid planning a security workflow around immediate broad access. The most sensible approach is to monitor Google’s official availability updates, evaluate access eligibility through Fairwind where appropriate, and keep existing code review, incident response, and vulnerability management procedures in place.
What do Gemini 4 Carbon’s benchmarks show?
Gemini 4 Carbon posted strong security benchmark results in Google’s testing, but the figures are self reported and do not establish independent real world performance. Google reports an 88.2% score on SEC-bench Pro for finding vulnerabilities, a 54.8% score on ExploitGym for exploitation, and a 71.0% score on CWE-bench for patching. The results indicate that Google tested Carbon across three distinct security tasks rather than measuring only code generation.
| Security task | Benchmark | Google reported result | What the result measures |
|---|---|---|---|
| Vulnerability discovery | SEC-bench Pro | 88.2% | Finding security weaknesses in benchmark tasks |
| Exploitation | ExploitGym | 54.8% | Completing exploitation focused tasks |
| Code patching | CWE-bench | 71.0% | Producing patches for known weakness categories |
Google also says Gemini 4 Carbon generated secure code at more than twice Gemini 4 Argon’s success rate in its internal benchmarks. That comparison may be useful for organizations already evaluating Gemini 4 models, but it remains a vendor measurement with no independent methodology or external replication included in the available announcement.
Security teams should test Gemini 4 Carbon against their own codebase, policies, and threat model before using its outputs in a production process. A benchmark can show capability under a defined evaluation, but it does not confirm that a model understands an organization’s architecture, deployment constraints, or acceptable operational risk.
How much does Gemini 4 Carbon cost through the API?
Gemini 4 Carbon costs $2 per million input tokens and $10 per million output tokens during Google’s introductory API pricing period. Cached input receives a 95% discount, according to Google. Input tokens generally represent material sent to the model, while output tokens represent the model’s generated response, so total cost depends on both prompt volume and response length.
Gemini 4 Carbon pricing is set to rise after the introductory period to $4 per million input tokens and $20 per million output tokens. Google did not provide a date for that change in the available announcement. The price increase matters most for teams using large repositories, long logs, extensive code context, or repeated automated workflows.
Organizations should estimate costs using the later rates rather than only the introductory figures if they are designing an ongoing security program. Cached input can reduce the cost of repeatedly analyzing stable documentation or shared code context, but teams should confirm current billing terms before deployment because Google has not provided a public end date for the introductory period.
How does Gemini 4 Carbon compare with Gemini 4 Argon?
Gemini 4 Carbon and Gemini 4 Argon are separate Gemini 4 models with different stated priorities and rollout paths. Google positions Carbon for cybersecurity and long horizon engineering tasks, while Gemini 4 Argon launched earlier and has been rolling out to paying subscribers. Reuters reported that Carbon arrived two weeks after Argon as Google competes in coding and cybersecurity AI. Reuters’ launch report provides the broader timing and competitive context.
| Model | Google’s stated focus | Availability described in current reporting |
|---|---|---|
| Gemini 4 Carbon | Cybersecurity and long horizon engineering | Selected defenders and trusted Fairwind partners initially |
| Gemini 4 Argon | Separate Gemini 4 model release | Global rollout to paying subscribers |
Gemini 4 Carbon may be the more relevant model for a security team evaluating vulnerability discovery, exploitation research, or patching assistance. Gemini 4 Argon may remain the more accessible option for users who already have qualifying paid access. The practical choice depends on whether an organization needs specialized security evaluation or broad product availability.
What safety claims has Google made about Gemini 4 Carbon?
Gemini 4 Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0% and reduced high severity misalignment alerts by almost 80% in Google’s internal live coding environments. Google presents those results as evidence that the model follows intended task boundaries more reliably during coding evaluations. The figures are self reported internal results, so outside researchers have not independently verified their scope or durability.
Gemini 4 Carbon’s safety claims matter because cybersecurity models can help defenders but can also create risk when they provide incorrect code, unsafe remediation advice, or high confidence output that has not been checked. Security teams should preserve human approval for vulnerability disclosures, production patches, access changes, and any action that could affect customer systems or sensitive data.
Organizations should also avoid placing credentials, private keys, customer records, or sensitive incident data into an AI prompt unless their approved agreement and data handling controls specifically permit that use. Teams evaluating related AI access tools should also consider how permissions expand the data available to a model, as shown by Gemini’s app and skill permissions.
When will Gemini 4 Carbon reach consumers and businesses?
Gemini 4 Carbon will reach developers, businesses, and consumers through Google product surfaces “as soon as possible,” but Google has not announced a public availability date. The current statement confirms Google’s intention to expand access without confirming when a specific user, subscription tier, country, or API account will receive the model.
Gemini 4 Carbon’s staged release is particularly important because security tools can affect code, infrastructure, and incident handling. Limited access gives Google an opportunity to gather feedback from cybersecurity defenders before the model reaches a wider audience, but the announcement does not specify testing milestones or conditions that would trigger the next phase.
Developers and businesses should not rely on informal screenshots or third party claims of access. Check official Google product notices and the Fairwind Program information, then test any newly available model in a controlled environment before connecting it to source repositories or operational systems. Organizations facing an active incident should continue to use established security response procedures and contact qualified incident response professionals when necessary.
FAQ
Is Gemini 4 Carbon publicly available?
Gemini 4 Carbon is not broadly publicly available yet. Google has initially limited access to a subset of cybersecurity defenders and trusted Fairwind Program partners.
What is Gemini 4 Carbon designed to do?
Gemini 4 Carbon is designed for cybersecurity and long horizon engineering tasks. Google highlights vulnerability discovery, exploitation work, code patching, and secure code generation as core areas.
How much does Gemini 4 Carbon cost?
Gemini 4 Carbon initially costs $2 per million input tokens and $10 per million output tokens through the API. Google says prices will later increase to $4 for input and $20 for output per million tokens.
Are Gemini 4 Carbon benchmark results independently verified?
No, Gemini 4 Carbon’s published benchmark results are Google’s self reported measurements. Organizations should evaluate the model on their own security tasks before relying on its findings or code changes.
Can Gemini 4 Carbon replace cybersecurity professionals?
No, Gemini 4 Carbon should not replace cybersecurity professionals or established security review processes. Human experts should validate vulnerability findings, code patches, and any action affecting production systems or sensitive data.