On this page · 4 sections ▾
Google announced Gemini 4 Carbon on October 28, 2026, its second frontier model in about two weeks after Gemini 4 Argon. According to 9to5Google, Google reports significant gains in coding, agentic tasks and cybersecurity defense. Access starts with a select group of cyber defenders, with a wider rollout to follow.
Key facts
- Carbon ranks #1 on v5.0 of the Artificial Analysis Intelligence Index with a score of 59, per Google.
- Cybersecurity scores cited by Google: 88.2% on SEC-bench Pro (finding vulnerabilities), 54.8% on ExploitGym (exploitation) and 71.0% on CWE-bench (patching).
- On WebDev Arena, Carbon gained 89 Elo points to reach #4 at 1769.
- Introductory API pricing is $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20. Cached input tokens are 95% cheaper.
- Initial access is limited to cyber defenders through Google's Fairwind program.
What Google says changed
Google says Carbon was tuned using feedback from the Argon launch, with work on coding, agentic and scientific tasks, full-stack web development and multimodal understanding. The Verge reports that internal teams saw it run for days to push through dead ends on multi-step engineering tasks.
On reliability, Google claims Carbon cuts unsupported claims by roughly 2x versus Argon. It also says overreach fell 26% and under-service fell 29%. Android Authority adds that Google says Carbon more than doubles the success rate for generating secure code compared with Argon.
Ars Technica reports that in the Grey Swan evaluation Carbon resisted every prompt-injection attempt, a 0 percent score against 0.7 percent for Argon, 1 percent for Claude Opus 5.5 and 8.5 percent for GPT-6 Astra. All of these figures come from Google's own materials and internal testing.
How to read the benchmarks
Ars Technica urges caution. It notes that Carbon beats Argon by slim margins in every test Google published, and that it is unclear whether Carbon is a wholly new model or a post-training upgrade of Argon. It also says every frontier model scores in the same neighborhood, with Carbon ahead of GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1 by only a few points.
Digital Trends adds that Carbon still trails rivals on a couple of specific coding benchmarks. The pattern is a narrow lead rather than a clear break from competitors.
Release and pricing: what to watch
Google is following the US government's voluntary testing framework, which Ars Technica and Android Authority both cite as the reason for a staged release. Google has not promised a date for general availability. Ars Technica notes that Argon reached wider users in about two weeks, and suggests Carbon could follow a similar path.
Pricing is also open-ended. Argon's discounted rates run through the end of November, but Google has not said when Carbon's introductory rate ends, according to Ars Technica, and declined to give an end date when asked. Developers budgeting around the $2 and $10 rates should plan for the $4 and $20 rates reported by 9to5Google and Digital Trends.
