![Gemini 4 Carbon: $2 Intro Price, Not Yet Worldwide [2026]](https://shattered.io/wp-content/uploads/2026/11/gemini-4-carbon-availability-2-dollar-intro-price-not-worldwide-2026-gen.webp)
The headline making the rounds today says Google’s Gemini 4 Carbon is now widely available around the world. The record says otherwise. Google’s own announcement describes a staged rollout that began on October 28, 2026 with a subset of cyber defenders, with access expanding gradually to more trusted partners through the company’s Fairwind Program.
That gap matters for anyone planning a budget or a security roadmap around this model. Gemini 4 Carbon is real, it has a published API price, and Google calls it its strongest model for cybersecurity and long-horizon work. Open worldwide access, though, is not something any source we reviewed confirms. If you want the model-versus-model angle first, our Claude vs Gemini breakdown covers how Carbon and Argon stack up against Anthropic’s lineup. This piece stays on the availability question and what it means for buyers.
Don't miss new tech stories on Google
Add Shattered once in the Google app and our stories show up in your news suggestions.
What Google Announced on October 28, 2026 #
Google introduced Gemini 4 Carbon on October 28, 2026. In its own wording, the company calls it “our next frontier model” and says it is its “strongest model with state-of-the-art cybersecurity and long horizon capabilities.” Those two phrases carry the whole pitch. Carbon is not framed as a general chatbot upgrade. It is framed as a model for hard, long-running work, with security as the headline use case.
Press coverage followed the same line. Engadget’s report framed the release as a significant leap in AI coding and cybersecurity, which matches Google’s emphasis. Digital Trends described it as built for complex coding and cybersecurity tasks. Both outlets describe the launch itself. Neither says every user on Earth can open a console today and call the model.
The announcement also set a price. Google listed an introductory API rate of $2 per million input tokens and $10 per million output tokens, plus a 95% discount on cached input tokens. Later reporting lists a higher rate of $4 per million input tokens and $20 per million output tokens. We unpack both numbers in the pricing section below, because the difference between them will drive real bills.
One more detail from Google’s materials sets the tone. The company contrasts Carbon with its sibling model, Argon. Google said Argon remains the ideal choice for the majority of software engineering and knowledge work, while Carbon is built for tasks that test the limits of a model’s intelligence. That sentence tells you who the intended customer is. It is the team with a problem Argon cannot solve, not the team that wants a faster autocomplete.
Is Gemini 4 Carbon Available Worldwide? What the Sources Say #
Short answer: not confirmed. The materials Google published describe restricted testing and a gradual rollout. They describe wider access for developers, businesses, and consumers as something that is coming, and they give no date for it. The 9to5Google announcement coverage reports on the same launch, and none of the coverage we reviewed describes a global switch-on.
It helps to separate three states that headlines often blur together. A model can be announced, meaning the company has said it exists. It can be accessible to a limited group, meaning named testers or partners can use it. And it can be generally available, meaning anyone who signs up can use it in the standard products. Gemini 4 Carbon sits in the first two states. The third is the one the viral headline claims, and it is the one nobody has confirmed.
Why would the claim spread anyway? Part of the reason is the published price. When a vendor lists per-token rates, readers assume a public API exists. Another part is that Argon sits alongside Carbon, so assumptions about one bleed into the other. A third part is plain headline compression. “Rolling out” becomes “available” becomes “available worldwide” after two or three rewrites.
For practical purposes, treat Gemini 4 Carbon as an invitation-stage product. If you are not a cyber defender or a trusted partner in the Fairwind Program, plan on waiting for Google to announce the next stage. We will update this article when the company publishes a general-availability date. Until then, any page that says “available now to everyone” is running ahead of the evidence.
The Fairwind Program: Who Gets In First #
Google says initial access went to a subset of cyber defenders. From there, access expands gradually to additional trusted partners through its Fairwind Program. The company has not published, in the material we reviewed, a full list of participants or a timetable for each wave. That silence is normal for a staged launch, but it leaves outside teams guessing about when they might qualify.
Why a security model launches to defenders first
A model that is strong at security work is strong in two directions. The same skill that helps a defender find and fix a flaw can help an attacker find one. Gating early access to defenders gives the vendor time to watch real usage, tune safeguards, and learn how the model behaves before opening the doors. Google has not spelled out its internal reasoning in the sources we reviewed, so this is our reading of the incentives, not a Google statement.
What “trusted partner” probably means for you
Expect the early cohort to skew toward security vendors, large enterprises with mature security teams, and organizations that already have deep contracts with Google. Small teams and individual developers should not expect an invitation soon. If you run a security product, the practical move is to contact your Google account team and ask directly how Fairwind admissions work, rather than waiting for a public sign-up page that may not arrive for a while.
Gemini 4 Carbon Pricing: $2 Intro Rate vs $4 Listed #
Google’s announcement gives an introductory API price of $2 per million input tokens and $10 per million output tokens. Cached input tokens get a 95% discount off the input price. Later listings reported by outlets show $4 per million input tokens and $20 per million output tokens. Both pairs keep a 5-to-1 ratio between output and input cost, so the later rate is exactly double the introductory one.
The table below turns those rates into dollars. Rows marked “derived” are our own arithmetic from the published figures, not numbers Google printed. The workload rows use an illustrative month of 50 million input tokens and 5 million output tokens, with the cached case assuming 80% of input tokens hit the cache.
| Item | Introductory | Later listed | Change |
|---|---|---|---|
| Input, per 1M tokens | $2 | $4 | 2x |
| Output, per 1M tokens | $10 | $20 | 2x |
| Cached input, per 1M tokens (derived, 95% off) | $0.10 | $0.20 | 2x |
| Output-to-input price ratio | 5 to 1 | 5 to 1 | Unchanged |
| Illustrative month, no caching (50M in, 5M out, derived) | $150 | $300 | +$150 |
| Same month, 80% of input cached (derived) | $74 | $148 | +$74 |
Two things stand out. First, the cache discount is large enough to cut that sample workload roughly in half, from $150 to $74 at the introductory rate. Second, the move from introductory to listed pricing doubles every line. A team that signs a budget on the $2 and $10 figures and then scales up could see its bill double without changing a single prompt. We could not confirm from the sources what date the introductory rate ends, so build both scenarios into your plan.
Cost Math: What 95% Cached-Input Savings Do to a Real Bill #
Caching rewards a specific kind of workload. If your agent re-sends the same long system prompt, the same repository context, or the same policy document on every call, most of your input tokens repeat. Those repeated tokens are what the 95% discount targets. Long-horizon agents, which Google names as a Carbon strength, tend to carry large context from step to step, so they are natural beneficiaries.
The snippet below reproduces the table’s arithmetic so you can plug in your own volumes. It assumes the cache discount applies to the full input price and ignores any storage or write fees, because the announcement material we reviewed does not describe them. Check Google’s current pricing page before you trust the output for a contract.
def monthly_cost(m_in, m_out, cached_share, in_price, out_price):
cached = m_in * cached_share * in_price * 0.05
fresh = m_in * (1 - cached_share) * in_price
return round(cached + fresh + m_out * out_price, 2)
print(monthly_cost(50, 5, 0.0, 2, 10)) # 150.0
print(monthly_cost(50, 5, 0.8, 2, 10)) # 74.0
print(monthly_cost(50, 5, 0.0, 4, 20)) # 300.0
print(monthly_cost(50, 5, 0.8, 4, 20)) # 148.0
Notice that output tokens dominate once caching kicks in. In the 80% cached case at the introductory rate, output costs $50 of the $74 total. Agents that think out loud, write long patches, or produce verbose reports will feel the $10 output rate far more than the $0.10 cached-input rate. If cost is a constraint, cap output length and ask for diffs instead of full files.
For a wider view of how hosted pricing compares with running your own weights, see our self-hosted LLM versus API cost comparison. The break-even logic there applies here too. Carbon only makes sense when the task value per call clearly exceeds the token bill.
Carbon vs Argon: Two Models, Two Jobs #
Google’s positioning is unusually direct. Argon stays the default for most software engineering and knowledge work. Carbon targets the hardest problems. That split gives buyers a clear decision rule, and it also hints at how Google expects to price and ration capacity. A model reserved for the limit-testing tasks does not need to serve every autocomplete request.
| Aspect | Argon | Carbon |
|---|---|---|
| Google’s framing | Ideal for most software engineering and knowledge work | Built for tasks that test the limits of a model’s intelligence |
| Headline strength | General-purpose work | Cybersecurity and long-horizon capabilities |
| Access today | Not covered in this article | Subset of cyber defenders, expanding via Fairwind |
| Announcement | Preceding model | October 28, 2026 |
| Published API price | Not covered in this article | $2 in / $10 out intro, $4 / $20 later listed |
The practical takeaway for engineering leads: do not migrate a working Argon pipeline to Carbon because the new name sounds better. Pick a narrow set of tasks where Argon fails or stalls, run those on Carbon once you have access, and measure the gain against the extra cost. Without that test, you pay double or more for work that did not need it.
Our earlier piece on Gemini 4’s launch timing and the Meta Muse comparison covered an earlier round of claims about Gemini 4 timing. The same caution applies today.
How Carbon Fits the Competitive Picture #
Frontier labs now ship tiered lineups, with a flagship for the hardest work and cheaper siblings for volume. Google’s Argon and Carbon split follows that pattern. What makes Carbon distinct is the explicit security focus and the choice to start with defenders. Rivals have their own premium tiers, and the public record on how they compare on security tasks is thin and mostly vendor-reported.
Independent testing is the missing piece. Vendors publish their own results, and buyers should treat those as marketing until a third party reproduces them. A useful reference point is outside evaluation of offensive-security skill, such as the results we covered in our look at an open model’s score on the Vals cyber-offense test. That kind of neutral benchmark is what will eventually settle whether Carbon’s security claims hold up in the field.
Also, price will shape adoption as much as capability. At $2 and $10 per million tokens, Carbon’s introductory rate is aggressive for a model Google calls its strongest. The later $4 and $20 figures are still a premium, but the question for each buyer is whether a single prevented incident or a single week of saved engineering time pays for the difference.
Fact Check: What Is Confirmed and What Is Not #
Because the viral headline blends confirmed and unconfirmed claims, a line-by-line check is worth the space. The table separates what the announcement and named outlets support from what remains open.
| Claim | Status | Basis |
|---|---|---|
| Google announced Gemini 4 Carbon on October 28, 2026 | Confirmed | Google’s announcement |
| Google calls it its strongest model for cybersecurity and long-horizon work | Confirmed (Google’s own description) | Google’s announcement |
| Initial access went to a subset of cyber defenders | Confirmed | Google’s announcement |
| Access expands to more trusted partners via Fairwind | Confirmed | Google’s announcement |
| Intro API price is $2 input and $10 output per 1M tokens | Confirmed | Google’s announcement |
| Cached input is 95% cheaper | Confirmed | Google’s announcement |
| Later listed price is $4 input and $20 output per 1M tokens | Reported | Outlet listings |
| Wider access for developers, businesses and consumers | Described as forthcoming, no date | Google’s materials |
| Gemini 4 Carbon is widely available around the world | Unconfirmed | Sources describe restricted testing |
The last row is the one that matters for today’s news cycle. A reader who sees “widely available” and tries to sign up may hit a wall. A procurement lead who writes it into a plan may build on a foundation that does not exist yet. The safe reading is that Carbon is announced, priced, and in limited hands.
Historical Context: Why Staged Releases Are Becoming the Norm #
Not long ago, a big model launch meant a blog post, a public API, and a rush of developers on day one. That pattern has shifted. As models have grown better at tasks with dual-use potential, labs have leaned on trusted-tester programs, partner-first rollouts, and tiered access. Google’s Fairwind Program is one expression of that trend, and it fits a broader industry habit of releasing sensitive capabilities in steps.
Staged rollouts also serve commercial goals. Limited access lets a vendor control demand while capacity is tight, collect feedback from high-value customers, and shape the narrative before critics and competitors pile in. A restricted launch with a public price list is a way to signal confidence and to anchor expectations at the same time. None of this is unique to Google. It is how premium AI products increasingly reach the market.
The cost of that approach is confusion. Readers see a price and assume availability. Aggregator sites repeat the claim. Social posts strip out the word “gradual.” That is how a restricted rollout turns into a “widely available” headline within a week. Journalists and buyers alike have to read the fine print on who can actually call the model.
Market Impact: Developers, Security Teams, and Cloud Buyers #
For security teams, the immediate impact is limited to those inside the early cohort. Defenders with access can test Carbon against their own codebases and incident workflows. Everyone else watches and waits. The real market effect will show up when access widens, because that is when pricing, rate limits, and enterprise terms become negotiable facts rather than footnotes.
For application developers, the most visible effect is on planning. Teams that had been waiting for a stronger Google model now have a named product and a price to model against. Even without access, they can estimate costs, design their caching strategy, and decide which workloads would justify a premium tier. The Lindy explainer on who can use Gemini 4 Carbon is one third-party guide that walks through access and pricing questions in plain terms, though you should confirm any specifics against Google’s own pages.
For cloud buyers, the key signal is Google’s willingness to price a frontier model at $2 and $10 per million tokens, even if only for an introductory window. That sets a reference that competitors will be measured against in sales calls. Expect procurement teams to quote it back at rival vendors. Whether the later $4 and $20 listing holds will matter more for long-term contracts than the launch rate does.
The broader category page for AI and machine learning coverage keeps running updates on how these pricing and access moves play out across vendors.
Risks and Open Questions #
Several things remain unknown from the material we reviewed. Google has not given a general-availability date. It has not published a full participant list for Fairwind. It has not, in the sources we saw, said how long the introductory price lasts or which regions will get access first. Each gap is a planning risk for any team that wants to depend on the model.
Security risk deserves its own line. A model sold on finding and fixing vulnerabilities will attract attention from attackers who want to misuse it, and from researchers who want to test its limits. Models used in agent workflows also face prompt injection, where hostile text inside a document or web page steers the model. If you plan to put any frontier model into an agent that reads untrusted content, run your own checks first. Our prompt injection detector test shows a repeatable way to score defenses before you ship.
Finally, there is vendor-claim risk. Google describes Carbon as state-of-the-art in cybersecurity. That is Google’s description, and independent verification will take time. Until neutral testers publish results, treat the claim as a hypothesis worth testing, not a settled fact.
What to Do This Week #
If you lead a security or platform team, a short checklist beats waiting. First, ask your Google account contact whether you qualify for the Fairwind Program and what the admission process looks like. Second, build a cost model with both price pairs, $2 and $10 as well as $4 and $20, so a repricing does not surprise finance. Third, identify two or three tasks where Argon or another model currently falls short, and write them up as test cases for when access arrives.
If you run a content or marketing team, the lesson is about sourcing. Before repeating that a model is “available,” check the vendor’s own post for the access language. Words like “subset,” “gradually,” and “trusted partners” are signals that general access has not started. Quoting the access terms accurately protects your credibility and your readers’ time.
Predictions: Where Gemini 4 Carbon Goes Next #
These are our own analyst judgments, not statements from Google or any named person, and none should be read as confirmed plans.
- Access widens in steps, not all at once. Google’s language points to expansion through trusted partners first, so we expect more partner waves before any open sign-up.
- The introductory price will not be the long-run price. The later $4 and $20 listing suggests the $2 and $10 rate is a promotional window, and budgets should assume the higher figure.
- Caching becomes a purchasing criterion. With 95% off cached input, vendors that offer weaker caching terms will face pressure from buyers running long-horizon agents.
- Independent security benchmarks will decide the narrative. Vendor-reported results will matter less once neutral testers publish head-to-head numbers on vulnerability work.
- Argon stays the volume workhorse. Google’s own framing says Argon fits most engineering and knowledge work, so most production traffic will stay there even after Carbon opens up.
The Takeaway #
Gemini 4 Carbon is a real product with a clear target and a published price, announced on October 28, 2026. The claim that it is now widely available around the world is not supported by the sources that matter, starting with Google’s own announcement. Plan around a gradual rollout, model your costs at both price points, and keep an eye on Fairwind for the next sign of expansion.
FAQ: Gemini 4 Carbon Availability and Pricing #
Is Gemini 4 Carbon available worldwide right now?
No confirmation exists. Google’s materials describe initial access for a subset of cyber defenders and a gradual expansion to trusted partners. Wider availability for developers, businesses, and consumers is described as forthcoming without a date.
When did Google announce Gemini 4 Carbon?
Google announced it on October 28, 2026, describing it as its next frontier model and its strongest for cybersecurity and long-horizon work.
How much does the Gemini 4 Carbon API cost?
The introductory price is $2 per million input tokens and $10 per million output tokens. Outlets also list a later price of $4 per million input tokens and $20 per million output tokens. Cached input tokens get a 95% discount off the input price.
What is the Fairwind Program?
It is the channel Google uses to expand Carbon access to additional trusted partners after the first group of cyber defenders. Google has not published a public sign-up process in the material we reviewed.
What is the difference between Gemini 4 Carbon and Argon?
Google says Argon remains the ideal choice for most software engineering and knowledge work, while Carbon is built for tasks that test the limits of a model’s intelligence. Carbon also carries the headline emphasis on cybersecurity.
Should I switch my production workloads to Gemini 4 Carbon?
Not yet, and not wholesale. Most teams lack access, and Google itself points most work toward Argon. Test Carbon on a few hard tasks once you qualify, and compare the gain against the higher token cost.
Does the 95% cache discount apply to output tokens?
No. The announced discount applies to cached input tokens, priced at 95% off the input rate. Output tokens keep their listed price.