Gemini 4 Carbon: $2/$10 at launch, then $4/$20, but only cyber defenders get it for now

Gemini 4 Carbon is Google's newest Gemini model, announced on October 28, 2026, two weeks after Gemini 4 Argon went on general sale. Google calls it "our strongest model with state-of-the-art cybersecurity and long horizon capabilities." You probably can't use it yet. Google says Carbon goes first to "a subset of cyber defenders," then to more partners in its Fairwind Program, and only after that to developers, businesses and consumers, with no date. When it does launch, Google says it will cost the same introductory $2/$10 per million tokens as Argon, then $4/$20. Argon took the same path: it debuted for Fairwind partners only on September 30, as our Argon launch story reported, and reached paying users on October 13.

This story is based on Google's October 28 announcement, Google DeepMind's Gemini page and Carbon evaluation methodology, the Gemini API models, pricing and changelog pages, Google's model card index, plan pages and app release notes, Sundar Pichai's October 28 earnings remarks, Artificial Analysis's October 28 results and leaderboard, and Vals AI's Vals Index, checked October 29, 2026.

The short version

• Release date: announced October 28, 2026. Google says a subset of cyber defenders gets it "starting this week." There's no public release date.
• Price: $2 input and $10 output per million tokens at launch, with cached input 95% off; $4/$20 "after the introductory period expires." Google hasn't said when that is.
• Where you can use it: nowhere public yet. Carbon isn't in the Gemini API, AI Studio, the Gemini app or Google's plan pages as of October 29.
• Benchmarks: Artificial Analysis ranks Carbon (high) first on its Intelligence Index v5.0 at 59. Google's cybersecurity and coding numbers are vendor-reported.
• Not published: a model ID, a context window, a model card or a Carbon-specific safety report.

When is the Gemini 4 Carbon release date?

There isn't one for the public. Google's post says: "Starting this week, we will make Carbon available to a subset of cyber defenders," and that it "will gradually phase access to additional trusted partners through our Fairwind Program." Only after gathering feedback and working on guardrails will Google make Carbon available "to developers, businesses, and consumers via our product surfaces as soon as possible."

Fairwind is Google DeepMind's access program for vetted defenders. Its page says it prioritizes "governments and partners most critical to societal resilience," such as national cyber authorities, critical infrastructure operators and core technology platforms. It says it works with more than 650 partners and that organizations request access through an interest form. As of October 29, the Fairwind page still names only Argon.

For comparison, Argon spent 13 days in its Fairwind-only phase before general availability. Google hasn't said whether Carbon will follow the same timetable.

How much will Gemini 4 Carbon cost?

Google says Carbon "will launch at the same introductory price as Argon at $2 per million input tokens and $10 per million output tokens," with cached input "priced at 95% off input token price." A footnote adds: "After the introductory period expires, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply."

Google gives no end date for Carbon's introductory period. Argon's has one: the Gemini API pricing page lists Argon at $2/$10 "through Nov 30, 2026" and $4/$20 from December 1, with no free tier. Carbon itself doesn't appear on the pricing page yet, so there are no batch, caching or free-tier details. Artificial Analysis says Google is offering "a 50% launch discount, with no confirmed end date." There's no word on which Google AI subscription plans will include Carbon; Google's plan pages still list Argon as the top model. Our guide to Argon's price and plans covers what you can buy as of October 29.

What has Google published about Carbon, and what hasn't it?

Last verified: October 29, 2026

For reference, Argon's API limit is 1,048,576 tokens shared by input and output, per its model page. Whether Carbon matches that isn't stated by Google.

What benchmarks does Google claim for Gemini 4 Carbon?

These are vendor-reported: Google picked the benchmarks and ran Carbon "with the highest thinking settings," single attempt, and mostly took rivals' scores from those rivals' own reports. Google says the external cybersecurity results were "evaluated without guardrails."

• SEC-bench Pro (finding vulnerabilities): 88.2%, against 83.4% for Argon and 85.4% for GPT-6 Astra in Google's table. Google's methodology says each of the 183 tasks was averaged over three attempts.
• ExploitGym (writing exploits): 54.8%, against 42.0% for Argon. The methodology reports both a 6-hour and a 10-hour timeout and doesn't say which gives the headline score.
• CWE-bench (patching): 71.0%, against 68.0% for Argon and 67.0% for Claude Opus 5.5.
• GraphWalks, 256K to 1M tokens (long-context reasoning): 96.2%, against 84.2% for Argon. Google computed these itself.
• WebDev Arena: Google says Carbon gained 89 Elo points to reach fourth place at 1768.
• Prompt injection: Google says Carbon's attack success rate on Gray Swan's indirect prompt-injection benchmark is 0.0%, against 0.7% for Argon.

Google's own table doesn't put Carbon first everywhere. It shows Opus 5.5 ahead on OSWorld-2.1 computer use (87.2% vs 81.5%) and GPT-6 Astra ahead on FrontierSWE 2 coding (65.5% vs 61.2%). Google also cites a Chrome team report that Carbon is about 10 times better at generating good patches than "another lab's large frontier model," without naming it.

What do independent evaluators say?

Two independent evaluators have published Carbon results. Artificial Analysis scored Gemini 4 Carbon (high) 59 on its Intelligence Index v5.0 on October 28, which it calls "the new #1," 9 points above Argon; its public leaderboard listed Carbon first when checked on October 29. The index is a weighted composite of nine evaluations the firm picks; v5.0 launched on October 27, so its scores can't be compared with earlier versions. Artificial Analysis says Carbon "is being rolled out to selected users and is not yet publicly available," and doesn't say how it got access. It also reports Carbon's hallucination rate on its AA-Omniscience test rose to 11% from Argon's 4%, still the second-lowest it has measured. Vals AI's Vals Index v2.2, checked October 29, lists Carbon second at 70.36%, behind Claude Sonnet 5.5 at 70.85%. Our Intelligence Index v5 explainer covers what changed in the index, and our Carbon vs Opus 5.5 vs GPT-6.1 Sol comparison puts the numbers side by side.

How is Gemini 4 Carbon different from Argon?

Google presents Carbon as the next step in the same Gemini 4 series, not a replacement announced for sale. Its post says Carbon is "building on Gemini 4 Argon" and uses "the same multi-layered safety and security stack." On Alphabet's earnings call the same day, Sundar Pichai said: "Argon is just the start of the Gemini 4 series. Today, we announced Carbon," adding that Carbon "outperforms Argon across almost every domain."

The practical difference is access. Argon is on sale in the paid Gemini API, Google Cloud, the Gemini app for AI Pro and Ultra subscribers, and Antigravity (our how-to guide has the steps). Carbon is limited to vetted defenders. Google also claims better behavior than Argon in its internal tests, including about half as many unsupported claims and almost 80% fewer high-severity misalignment alerts in live coding environments. Google hasn't published a model card or safety report for Carbon to back those figures; its post links only the earlier Gemini 4 safety report. Google says it is "actively engaged in the U.S. government's voluntary process for pre-release model testing."

What's confirmed, reported and uncertain

• Confirmed (Google): the October 28 announcement; the cyber-defender and Fairwind rollout; the $2/$10 launch price, 95% cached-input discount and later $4/$20 price; that Carbon isn't yet on the Gemini API, plan or model card pages.
• Vendor-reported: every Google benchmark and behavior figure above, including the WebDev Arena rank and Gray Swan result.
• Third-party: the Intelligence Index v5.0 score of 59 and the hallucination figures (Artificial Analysis, October 28); the Vals Index v2.2 score of 70.36% (Vals AI, checked October 29).
• Uncertain: the public release date; the model ID; the context window (Artificial Analysis lists 1.0M tokens, Google hasn't said); when the introductory price ends; which Google AI plans will get Carbon; whether a model card will be published.