Google's newest model tops a major AI ranking, but security defenders get it before anyone else. Here is the rollout order, the price and how it stacks up.
Google announced Gemini 4 Carbon on Wednesday, its most capable AI model so far, but most people can't use it. Access is limited to vetted cybersecurity defenders in Google's Fairwind Program, and developers, businesses and consumers will follow later with no date set.
Going by Google's past releases, wider access could come within one to two weeks, but Google won't confirm a date. For Gemini app users, nothing changes yet.
Is Gemini 4 Carbon available? Not to you, not yet. Google announced its most capable AI model on Wednesday, then handed it only to a small group of vetted cybersecurity defenders, the people who hunt for and fix software flaws for a living.
That is an unusual way to launch a model that just took first place on a widely watched independent ranking. If you pay for Gemini, build apps on Google’s AI or simply want to know whether the chatbot on your phone is about to get smarter, here is who gets Carbon first, what it will cost, and how much better it really is than what you can use today.
Only selected security professionals in Google’s Fairwind Program have access today. Google says it will widen access step by step, first to more trusted partners and then to developers, businesses and everyday consumers. It has not given a date for any of those stages.
When asked directly, Google declined to say when access would open up. The company also says it is taking part in the US government’s voluntary review of models before release, and that wider availability depends on how its safeguards hold up in real-world use.
The best clue is Google’s last model. Gemini 4 Argon was released in stages too, and paying developers could use it from October 13, about a fortnight ahead of the Carbon announcement. If Google follows the same pattern, developer access could arrive within a week or two, though that is an estimate based on past releases, not a promise.
Consumers should expect to wait longer. Google has not said whether Carbon will replace Argon inside the Gemini app or which subscription tiers would get it first.
Carbon’s biggest gains are in finding and fixing security holes, and that skill cuts both ways. A model that can spot a vulnerability and prove it can be exploited is a gift to defenders, and a risk if attackers get it early.
Google’s own numbers show where the jump is:
Google also says Carbon is twice as likely as Argon to write code without vulnerabilities, and its Chrome team claims the model is 10 times better at producing high-quality patches than models from outside Google. Google ran the outside cyber tests with safety filters switched off, so the scores show raw ability rather than what a regular user would see. The pitch is that defenders can patch problems before the same capability spreads more widely.
Carbon leads, but only just. Carbon scored 59 on the Artificial Analysis Intelligence Index, a ranking run by an outside firm, Claude Opus 5.5 scored 58 and GPT-6 Astra scored 57. The public leaderboard had not yet listed Carbon when it was checked on Wednesday, and most head-to-head figures come from Google’s own table.
It doesn’t win everything. On the FrontierSWE 2 coding test, GPT-6 Astra reached 65.5% and Opus 62.3%, with Carbon at 61.2%. It did climb 89 points on the WebDev Arena leaderboard to fourth place. According to Google, its own engineers saw the model keep at a large codebase for days, working around dead ends until the bug was gone, and that it has cut back on inventing facts, taking shortcuts and doing things it wasn’t asked to do.
Is it really new, or a heavily tuned Argon? Google presents it as a new model, but outside observers note the two share a lot and Carbon beats Argon by small margins on each test.
Google charges for models like Carbon by the token, the small chunks of text a model reads and writes. Carbon’s launch price is $2 for every million input tokens and $10 per million output tokens. Google says those rates will later double to $4 and $20, without saying when.
That puts it above the cheapest rivals. OpenAI cut prices for GPT-6.1 Terra to $1 and $5 on October 26, so developers choosing a model for everyday tasks may find the top-ranked option isn’t the best value. For people who just use the Gemini app, token prices don’t apply; what matters is when Carbon reaches the app at all.
For most people, nothing changes this week. The Gemini you use today runs on older models, and there is no public sign-up for Carbon. If you’re a developer, plan around Argon for now and watch for Carbon in the API, ideally while the lower launch price still applies.
The bigger picture is a shift in how AI arrives. Argon was Google’s first new frontier model in 11 months; Carbon followed just two weeks later, and all three big labs have released new models this month. Staged launches that start with security teams may become the norm for the most capable systems.
No. Google hasn’t added it to the app and hasn’t said which plans will get it first.
It narrowly beats OpenAI’s GPT-6 Astra on the Artificial Analysis index and on hacking-related tests, but trails it on some coding benchmarks.
Right now only vetted defenders in the Fairwind Program can. Everyone else has to wait for Google to open the next stage.
You're reading the quick version.
Read next
Genie 4 waitlist: how to try Google’s AI worlds, and the catch
Google's new world model builds explorable scenes with sound in real time. Here is who gets it first, what the Ultra plan costs and what the waitlist means.
Google's new world model builds explorable scenes with sound in real time. Here is who gets it first, what the Ultra plan costs and what the waitlist means.
The first AI-made episode is free on X, but it falls far short of the full film Musk promised. What it shows, why critics hate it, what is next.
OpenAI says its internal models disproved a widely trusted computer science assumption and cracked open problems in math and physics. Experts now have to check the work.
The Daily Brief
Every morning, the news that matters from AI, phones, apps and the people shaping tech. Explained in plain English. Free.
One email a day. No spam, unsubscribe anytime. Privacy policy
