Google unveils Gemini 4 Carbon, scoring 88.2% on SEC-bench Pro, for cyber defenders first

What Google announced

Google DeepMind introduced Gemini 4 Carbon on 28 October 2026 in a post by Koray Kavukcuoglu, Google's chief AI architect. Google calls it its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. It follows Gemini 4 Argon, which launched two weeks earlier.

Google says Carbon scores 59 on version 5 of the Artificial Analysis Intelligence Index, which ranks it first. On cybersecurity benchmarks run without guardrails, Google reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching. Google also says Carbon gained 89 Elo points on WebDev Arena to reach fourth place at 1768. These are vendor-reported figures.

Who gets access and what it costs

Carbon is not generally available. Google says it will give access this week to a subset of cyber defenders and then widen access to trusted partners through its Fairwind Program. That program gives governments, healthcare providers and telecom operators early access to models for defensive work, and Google says it has over 650 partners. Google also says it is taking part in the US government's voluntary process for pre-release model testing.

Developers, businesses and consumers will get Carbon later, with no date given. Google says Carbon will launch at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. After the introductory period the price rises to $4 and $20. The context window, output limit and API model name are not stated in the announcement.

Safety claims and what changed from Argon

Google says Carbon uses the same safety stack as Argon, with training, real-time misuse defenses, prompt-injection resilience and chain-of-thought monitoring. According to Google, Carbon cut successful cheating on Terminal-Bench 4.0 to 0.0%, reduced unsupported claims by about half compared with Argon, and resisted every attack in Gray Swan's indirect prompt injection benchmark, where Argon had a 0.7% attack success rate. Google says high-severity misalignment alerts in internal coding use fell by almost 80%.

The main change is the release order. Argon went straight to the Gemini API and the Gemini app at launch, as covered in our Argon release story. Carbon goes to security teams first. That mirrors how labs have handled models with strong cyber capabilities this autumn, including the limited release around Anthropic's exploit chain research.

What to watch next

Artificial Analysis, which tested the model, lists a 1M-token context window and text, image, video and speech input, but Google's post does not confirm those details. The open questions are the date of general availability, the API model name and whether Google will publish a model card or safety report for Carbon, as it did for Argon. Independent results on the Artificial Analysis index will be the first outside check on Google's ranking claim. The AI model release tracker will add API prices once Carbon reaches the Gemini API. More model news is on the AI Models hub.

Questions people ask

Can developers use Gemini 4 Carbon today?

No. Google says access starts with a subset of cyber defenders and Fairwind Program partners, and that developers, businesses and consumers will get it later. No date is stated.

How much will Gemini 4 Carbon cost?

Google says the introductory price is $2 per million input tokens and $10 per million output tokens, the same as Argon. After that period it becomes $4 and $20.

How does Gemini 4 Carbon score on cybersecurity benchmarks?

Google reports 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench, all evaluated without guardrails.

Sources

1. Gemini 4 Carbon: our next frontier model (Google, 28 Oct 2026)
2. Fairwind Program (Google DeepMind, 28 Oct 2026)
3. Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5 (Artificial Analysis, 28 Oct 2026)