Gemini 4 Carbon: cyber defenders only, no public API date
Gemini 4 Carbon tops Google's cyber benchmarks but ships only to vetted defenders, with no public API date. Plan on Argon and shorten your patch cycle.
Google has a new top model, and you cannot use it yet. Gemini 4 Carbon, announced October 28, is Google's strongest model for cybersecurity and long-running engineering work. It goes first to a small group of vetted cyber defenders. There is no date for the public Gemini API. If your roadmap has a line that says "switch to the newest Gemini," Carbon is not that line. Argon still is.
What actually happened
Per Google's announcement, written by Google DeepMind's Koray Kavukcuoglu:
- Access: a subset of cyber defenders gets Carbon starting this week. More trusted partners come in gradually through Google's Fairwind Program. Developers, businesses and consumers get it "as soon as possible," with no date.
- Cyber benchmarks (Google-reported): 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
- Long tasks: Google says internal teams watched Carbon work on hard bugs for days across large codebases.
- Pricing, when it opens: $2 per million input tokens and $10 per million output at the introductory rate, then $4/$20. Cached input is 95% off. That is the same price structure as Argon.
Alphabet's Q3 earnings release the same day puts Argon and Carbon side by side as Google's frontier pair, and says its first-party model APIs now process nearly 32 billion tokens per minute.
Why Gemini 4 Carbon matters for your business
Two things, and neither is "upgrade."
Do not plan around a model you cannot call. We said this about Argon before its GA, and it holds again. Your agents run on what is in the API today. Same price as Argon means no cost reason to wait, either. If you ran the Argon eval, keep the harness. When Carbon opens, you rerun the same 50 tasks in an afternoon.
The patch clock just got shorter. Google is gating Carbon because a model that scores 54.8% on exploitation is useful to attackers too. Defenders get it first. Attackers get the next open model that comes close. Old plugins, an unpatched WordPress install, a stale npm lockfile: these are the bugs a tireless model finds first. A monthly patch habit is no longer slow-and-fine. It is slow.
Key takeaways
- Gemini 4 Carbon launched October 28 for a subset of cyber defenders only
- No public API date; access widens through Google's Fairwind Program
- Google-reported scores: 88.2% SEC-bench Pro, 54.8% ExploitGym, 71.0% CWE-bench
- Pricing when it opens matches Argon: $2/$10 intro, then $4/$20 per million tokens
- Keep building on Argon, keep your eval harness, and patch your stack weekly
The best security model in the world does nothing for a plugin you forgot you installed. We audit small-business stacks, cut the dead dependencies, and set up update pipelines that run without you. Book a stack review.
Sources: Google: Gemini 4 Carbon, our next frontier model, Alphabet Q3 2026 earnings release.
- #gemini-4-carbon
- #ai-security
- #api-pricing
- #patch-management
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
OpenAI plans to buy Google TPUs: what it means for your AI bill
OpenAI plans to buy Google TPUs, per The Information and Bloomberg. No terms yet. Why AI chip supply shifts matter to your API costs and model routing.
Read itNvidia Nemotron 3.5 Transcribe: open speech-to-text, no streaming
Nvidia's Nemotron 3.5 Transcribe is an open-weight speech-to-text model for 40 languages and hour-long audio. It's batch-only. Here's where it fits.
Read it