Gemini 4 Carbon

Gemini 4 Carbon is the frontier model Google announced on 28 October 2026, two weeks after Gemini 4 Argon, in a post by Koray Kavukcuoglu, SVP of Google DeepMind and Google’s chief AI architect. Google calls it its strongest model and describes it as built for long-running agentic engineering work, such as root-causing and patching subtle bugs across large codebases, and for cybersecurity defence: finding vulnerabilities, checking whether they can be exploited, and writing the fix.

## Who can use it

At release Carbon is not generally available. Google says that from the week of 28 October it goes to a subset of cyber defenders, with access widened gradually to trusted partners through its Fairwind Program, and that it is taking part in the US government’s voluntary process for pre-release model testing. Developers, businesses and consumers are promised access “as soon as possible”, with no date. Artificial Analysis lists no public API for the model when read on 29 October. A buyer evaluating it today should plan on Argon, which Google’s related post says is now available to developers, and treat Carbon as a later upgrade.

## Price

Google gives the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. A footnote to the post says $4 and $20 apply once the introductory period ends, without saying when that is; the board records the later rate.

## What Google claims, and what was measured independently

Google’s post reports 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench, evaluated without guardrails, a doubling of its internal secure-code success rate over Argon, and 0.0% attack success on Gray Swan’s indirect prompt-injection benchmark. These are Google’s own evaluations and are not stored as scores here. Artificial Analysis, testing independently, scores Carbon 59 on version 5.0 of its Intelligence Index at its High setting, first of 52 models on its page when read on 29 October, against a median of 42; it lists text, image, speech and video input, text output and a 1 million-token context window. SiliconANGLE reported the release the same day. Google’s post gives no model identifier or knowledge cutoff.

## Key facts

**Maker:** Google (Google DeepMind) **Announced:** 28 October 2026 **Access:** cyber defenders and Fairwind Program partners first; general availability not dated **Context window:** 1 million tokens (per Artificial Analysis) **Price:** $2 in and $10 out per million tokens at launch, then $4 and $20 **Board family:** Gemini 4, after Gemini 4 Argon