Gemini 4 Carbon: What's New, Pricing, and Alternatives (2026)
Gemini 4 Carbon is Google's newest frontier model, announced on October 28, 2026. Google calls it "our strongest model with state-of-the-art cybersecurity and long horizon capabilities." It follows Gemini 4 Argon, which reached general availability in the Gemini API on October 13.
Carbon is not open to everyone yet. Google is starting with a small group of cyber defenders and plans a wider release later. This guide covers what changed, the benchmark results Google published, who can use it, what it will cost, and the main alternatives. Every detail comes from vendor pages read on November 2, 2026.
What Google announced on October 28
Google posted the announcement on its Keyword blog on October 28, 2026, a Wednesday. The post is signed by Koray Kavukcuoglu, SVP at Google DeepMind and Google's Chief AI Architect.
The headline claims are short. Google says Carbon "ranks #1 on the recently updated v5.0 of Artificial Analysis Intelligence Index at 59." It also reports large gains in coding, agentic tasks, and security work.
The post makes three points about the release itself:
- A staged rollout. Access starts with a subset of cyber defenders, then expands to trusted partners.
- A government review step. Google says it is "actively engaged in the U.S. government's voluntary process for pre-release model testing."
- Wider access later. Google says it will keep refining guardrails "before making Carbon available to developers, businesses, and consumers via our product surfaces as soon as possible."
So the model exists and has a published price, but most people cannot call it yet.
What is new in Gemini 4 Carbon
Google frames Carbon around three areas: long tasks, security, and safer behavior.
Long-horizon engineering. Google says "Carbon is built for sustained agentic workflows." Its teams have seen the model "run for days to persist through dead ends" on multi-step engineering tasks. The examples include root-causing and patching subtle bugs across large codebases.
Cybersecurity defense. The post covers the full defense cycle: finding flaws, checking which ones can be exploited, and fixing them. On internal tests, Carbon shows "more than 2x uplift in success rate for generating secure code over Argon."
Web development. Google reports better output in full-stack web work. Carbon gained 89 Elo points and reached fourth place on WebDev Arena, with a score of 1768.
Behavior and honesty. Several changes affect anyone who hands the model a real task:
- In internal testing, Carbon "cuts unsupported claims by approximately 2x over Argon."
- It "puts methodological caveats directly into the deliverable" instead of guessing around missing evidence.
- When a request is unclear, it "retrieves the relevant sources and surfaces discrepancies to the user before acting."
- It reduced overreach by 26% and under-service by 29%, according to Google.
These points matter for reports and analysis, not just code. A model that flags gaps is easier to check than one that fills them in.
The benchmark results Google published
Google lists the following results for Carbon. Google notes that the external cyber benchmarks were "evaluated without guardrails."
| Benchmark | What it measures | Carbon result |
|---|---|---|
| Artificial Analysis Intelligence Index v5.0 | Overall model capability | 59, ranked first |
| SEC-bench Pro | Finding real vulnerabilities | 88.2% |
| ExploitGym | Checking whether flaws can be exploited | 54.8% |
| CWE-bench (Collinear) | Patching flaws at the root cause | 71.0% |
| WebDev Arena | Web development output | 1768, fourth place |
| Gray Swan IPI Benchmark | Attack success rate from prompt injection | 0.0% (Argon: 0.7%) |
| Terminal-Bench 4.0 | Successful cheating on tasks | 0.0% (attempts: 0.4%) |
Two further claims come from Google's own teams rather than public benchmarks. Google says high-severity misalignment alerts in live coding "dropped by almost 80%." It also says the Chrome team saw Carbon perform "more than 10X better" than another lab's large model at writing patches.
Treat these as vendor figures. They are useful for direction, but they are not independent tests of your own tasks.
Who can use Gemini 4 Carbon today
Right now, very few people can use Gemini 4 Carbon. Google's post says access goes first "to a subset of cyber defenders." It then extends "to additional trusted partners through our Fairwind Program."
We checked the main access points on November 2, 2026:
- Gemini API. The model list and pricing page show Gemini 4 Argon as the newest model. Carbon does not appear yet.
- Gemini app. The Gemini Apps Help page on models lists Flash-Lite, Flash, Pro, and Argon. Carbon is not on the plan table yet.
- Google AI plans. The plans page promotes Gemini 4 Argon for Google AI Pro and Ultra. It does not mention Carbon.
Google's closing line sets expectations: "we look forward to bringing Carbon to developers, businesses, and consumers soon." No date was given.
Gemini 4 Carbon pricing
Google has published an API price before opening general access. The post lists Carbon's standard rate at $4 per 1M input tokens and $20 per 1M output tokens. Google says Carbon launches at the same price as Argon, and it also lists a lower rate for cached input tokens.
That matches Argon's standard paid-tier rate on the Gemini API pricing page: $4.00 per 1M input tokens and $20.00 per 1M output tokens. The pricing page also says Argon "is not available on the Free tier." Google has not yet said whether Carbon will follow the same rule.
For the Gemini app, Google has not said which plans will include Carbon. As a reference point, here is what the Google AI plans page showed on November 2:
| Google AI plan | Price shown on the plans page | Includes Gemini 4 Argon |
|---|---|---|
| Google AI Plus | $4.99 USD/mo | Not listed |
| Google AI Pro | $19.99 USD/mo | Yes |
| Google AI Ultra 5x | $99.99 USD/mo | Yes |
| Google AI Ultra 20x | $199.99 USD/mo | Yes |
The plans page shows these as monthly amounts. It does not say whether a yearly billing option exists for each plan. Check the page for your country before you subscribe, since prices and plans vary by region.
How Carbon compares with Gemini 4 Argon
Argon and Carbon are two weeks apart. Argon is the model you can use today. Carbon is the next step, with access still limited.
| Gemini 4 Argon | Gemini 4 Carbon | |
|---|---|---|
| Announced | Generally available in the Gemini API on October 13, 2026 | October 28, 2026 |
| Google's description | "our most capable model built for complex coding, agentic, math, and science tasks" | "our strongest model with state-of-the-art cybersecurity and long horizon capabilities" |
| Gemini API | Available on the paid tier | Not listed yet |
| Gemini app | Google AI Pro and Ultra | Not listed yet |
| Standard API rate | $4 input / $20 output per 1M tokens | $4 input / $20 output per 1M tokens |
| Prompt injection attack success | 0.7% (Gray Swan IPI) | 0.0% (Gray Swan IPI) |
If you need a frontier Gemini model this week, Argon is the practical choice. Our Gemini 4 Argon guide covers how it works in the Gemini app.
What Carbon means for everyday work
Most of Google's examples are about engineering and security. Even so, three changes carry over to everyday knowledge work once Carbon reaches the Gemini app.
Longer tasks without hand-holding. A model built to run for days on code should also hold up on long research or reporting jobs.
Clearer caveats. Google says Carbon writes caveats into the deliverable itself. For a market summary or a board memo, that means fewer hidden guesses to catch.
Questions before action. Carbon is trained to surface discrepancies before it acts. That can slow a task down slightly, but it reduces rework.
Until then, everyday users should plan around Argon and the models already in their tools.
How to prepare for Gemini 4 Carbon access
You cannot switch to Gemini 4 Carbon today, but you can get ready for it. A little preparation makes the first week of access far more useful.
Build a small test set now. Pick five to ten real tasks you run often. Include one long task, one task with messy inputs, and one task where accuracy matters. Run them on Argon today and save the results.
Write down what "good" looks like. For each task, note the facts the answer must include and the mistakes you see most. This gives you a fair way to judge Carbon against Argon later.
Check your plan and API tier. Argon is not on the Gemini API free tier, and Google has not said whether Gemini 4 Carbon will be. In the Gemini app, Argon sits on Google AI Pro and Ultra. Budget on the assumption that Carbon will need a paid plan too.
Watch the official pages. The Gemini API release notes and the Gemini Apps Help model page are where availability changes show up first. Both were unchanged for Carbon on November 2.
Keep your data workflow separate. If most of your work is reports and decks from spreadsheets, a new model will not change that workflow. Keep the tool you use for it stable while you test the model.
Gemini 4 Carbon alternatives
Since Carbon is not broadly available, these are the options you can use now. Prices come from each vendor's own page.
| Tool or model | What the vendor says it is for | Price, per the vendor |
|---|---|---|
| Gemini 4 Argon | "complex coding, agentic, math, and science tasks" | API: $4 input / $20 output per 1M tokens; app: Google AI Pro $19.99 USD/mo |
| Claude Opus 5.5 | "Daily driver for agentic coding and enterprise work" | API: $4 input / $20 output per MTok; Claude Pro $17 per month billed annually |
| GPT-6.1 Terra in Codex | Coding and computer tasks in OpenAI's Codex | Codex Plus $20/month; Pro from $100/month |
| Powerdrill Bloom | Reports, charts, and slides built from data files | Free plan; Pro $13.27/mo |
Gemini 4 Argon is the closest match. It shares Carbon's standard API rate and is already in the Gemini app on Google AI Pro and Ultra.
Claude Opus 5.5 has the same standard API rate as Carbon. Anthropic lists it as a "Daily driver for agentic coding and enterprise work."
GPT-6.1 Terra began "rolling out in ChatGPT Work and Codex" on October 26, according to OpenAI's release notes. Our GPT-6.1 Terra guide covers its computer-use features.
For a model comparison across vendors, see our Claude Sonnet 5.5 and Haiku 5.5 guide.
Where Powerdrill Bloom fits
Carbon is a model. You reach it through an API or an app, and you still shape the output yourself. Powerdrill Bloom works at a different layer. It takes your files and returns finished analysis, reports, and slides.
Its homepage says: "Every number comes back with the page, the row and the figure behind it." That fits the same goal Google describes for Carbon, which is fewer unsupported claims. The AI Report Generator builds reports from Excel, CSV, or PDF files. The Excel to PPT tool turns spreadsheet data into slides.
The two can sit side by side. Use a frontier model like Carbon for code and long technical tasks. Use Powerdrill Bloom when the output is a report or deck that must match the numbers.
On price, the pricing page lists a Free plan at $0.00 and Pro at $13.27/mo. To test it on your own data, try Powerdrill Bloom with one spreadsheet you report on every month.
Frequently asked questions
When was Gemini 4 Carbon released?
Google announced Gemini 4 Carbon on October 28, 2026. Access started that week for a subset of cyber defenders through Google's Fairwind Program. Google says it will bring Carbon to developers, businesses, and consumers soon, but has not given a date.
How much does Gemini 4 Carbon cost?
Google lists Carbon's standard API rate at $4 per 1M input tokens and $20 per 1M output tokens, the same as Gemini 4 Argon. Google also lists a lower rate for cached input tokens. It has not said which Gemini app plans will include Carbon.
Can I use Gemini 4 Carbon in the Gemini app?
Not yet. On November 2, 2026, the Gemini Apps Help page listed Flash-Lite, Flash, Pro, and Argon, but not Carbon. Google says consumer access will come after it gathers feedback and refines guardrails.
Is Gemini 4 Carbon better than Gemini 4 Argon?
Google says Carbon is its strongest model so far. It reports more than twice Argon's success rate at generating secure code in internal tests. Carbon also cut unsupported claims by about half compared with Argon. Argon remains the model you can use today.
What is the Fairwind Program?
Fairwind is the program Google is using to roll Carbon out in stages. Google says access starts with a subset of cyber defenders, then extends gradually to additional trusted partners. The post does not explain how to apply.
Sources: Google, Gemini 4 Carbon: our next frontier model · Gemini API release notes · Gemini Developer API pricing · Gemini API models · Gemini Apps Help, About Gemini models · Google AI plans · Claude pricing · Codex pricing · ChatGPT release notes. Details read from vendor pages on November 2, 2026.