Google announced Gemini 4 Carbon on October 28, 2026. It took first place on the Artificial Analysis Intelligence Index, but most people cannot use it yet. These answers are based on Google's announcement and the Artificial Analysis leaderboard as of November 2, 2026. Vendor-reported figures are marked as such.
Access and Price
Q: What is Gemini 4 Carbon?
It is Google's most capable Gemini 4 model, announced October 28, 2026, two weeks after Gemini 4 Argon became generally available. Google positions it for cyber defence and long-running engineering work.
Q: Who can use Gemini 4 Carbon?
At launch, "a subset of cyber defenders", followed by other trusted partners through Google's Fairwind Program. Google says developers, businesses and consumers will get it "as soon as possible", without a date. It is not in the Gemini app, AI Mode or the public API yet.
Q: What is the Fairwind Program?
It is Google DeepMind's trusted-access program for releasing security-relevant models to vetted defenders before a wider launch. Argon went through it first in September.
Q: How much does Gemini 4 Carbon cost?
Google lists an introductory $2 per million input tokens and $10 per million output tokens, with cached input 95% off, rising later to $4 / $20. Google has not said when the introductory price ends.
Performance
Q: How does Gemini 4 Carbon score on benchmarks?
On the Artificial Analysis Intelligence Index v5.0 it scores 59 at high effort and 58 at medium, first overall. Google also reports 88.2% on SEC-bench Pro, 71.0% on CWE-bench, 54.8% on ExploitGym, a 1768 Elo on WebDev Arena (fourth) and a 0.0% attack success rate on Gray Swan's indirect prompt injection test. Those last figures are Google's own.
Q: Is Gemini 4 Carbon better than Claude Opus 5.5 and GPT-6 Astra?
Slightly, on the Artificial Analysis index: 59 against 58 for Opus 5.5 (max) and 57 for GPT-6 Astra (max). That gap is smaller than the difference between effort settings on one model. See the full comparison.
Q: Does Gemini 4 Carbon hallucinate less than Argon?
Google says it makes about half as many unsupported claims. One secondary report cites an AA-Omniscience hallucination rate of 11% for Carbon against 4% for Argon, the opposite direction. We could not confirm that figure on Artificial Analysis' own page, so treat it as unverified until it is.
Q: What is the context window?
Artificial Analysis lists Carbon at 1M tokens, the same as Argon. Google hasn't published a full public API specification for Carbon.
For Brands
Q: Will Gemini 4 Carbon change how Gemini describes my brand?
Not yet, because consumers can't use it. When Google brings Carbon or its successors to the Gemini app and AI Mode, answers there will shift. Argon is the model to test today.
Q: Should I test my brand on Gemini 4 Carbon?
Only if you have access. Otherwise run your prompt set on Argon now, keep it as a baseline, and re-run when Carbon reaches public surfaces. See the Gemini 4 release brief.