Last updated: October 31, 2026
Gemini 4 Carbon is Google’s new frontier model, announced October 28, 2026, for long-horizon coding, agents and cybersecurity defense. It beats Gemini 4 Argon on every benchmark Google lists. Access is limited to selected cyber defenders and Fairwind partners for now, with launch pricing of $2 input and $10 output per million tokens.
This guide is based on Google’s Gemini 4 Carbon announcement, the Google DeepMind Gemini models page, the Fairwind Program page, and the Gemini API pricing and release notes pages, all checked on October 31, 2026. If you are new to Gemini 4, start with our Gemini 4 Argon power-user guide.
Contents
- What is new in Gemini 4 Carbon
- Gemini 4 Carbon vs Argon at a glance
- Gemini 4 Carbon benchmarks
- Who can use it and the Fairwind Program
- How to prepare for Gemini 4 Carbon
- Workflows that suit Carbon
- Limitations
- Frequently asked questions
What is new in Gemini 4 Carbon
In the announcement, Koray Kavukcuoglu of Google DeepMind describes Carbon as Google’s strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. It arrived about two weeks after Argon. The main changes Google describes are:
- Long-horizon engineering. Google says its teams have seen Carbon run for days on multi-step tasks, such as finding the root cause of subtle bugs across large codebases and low-level systems, and patching them.
- Cybersecurity defense. Google reports state-of-the-art results for finding vulnerabilities, checking whether they can be exploited, and patching them. On internal tests, it says Carbon has more than twice Argon’s success rate at generating secure code.
- Web development. Google says Carbon gained 89 Elo points on WebDev Arena, reaching fourth place at 1768.
- Alignment and honesty. Google says Carbon makes about half as many unsupported claims as Argon. It also reports 26% less overreach, 29% less under-service and a 0.0% attack success rate on Gray Swan’s indirect prompt injection benchmark, down from 0.7% for Argon.
Gemini 4 Carbon vs Argon at a glance
| Item | Gemini 4 Carbon | Gemini 4 Argon |
|---|---|---|
| Announced | October 28, 2026 | Generally available October 13, 2026 |
| Who can use it | A subset of cyber defenders, then Fairwind partners | Gemini API paid tier and Google AI Studio |
| API model ID | Not yet published | gemini-4-argon |
| Input price (per 1M tokens) | $2 introductory, then $4 | $2 through November 30, 2026, then $4 |
| Output price (per 1M tokens) | $10 introductory, then $20 | $10 through November 30, 2026, then $20 |
| Context window | Not yet published | 1M tokens, per the Gemini API models page |
| AA Intelligence Index v5.0 | 59 | 50 |
Argon details come from the Gemini API pricing page and release notes. Google’s Carbon post gives the same introductory and later prices as Argon, but does not say when Carbon’s introductory period ends.
Gemini 4 Carbon benchmarks
The DeepMind Gemini page publishes a comparison table. A selection is below. All figures are Google’s, and the blog notes that the external cybersecurity tests were run without guardrails.
| Benchmark | Carbon | Argon | GPT-6 Astra | Claude Opus 5.5 |
|---|---|---|---|---|
| AA Intelligence Index v5.0 | 59 | 50 | 57 | 58 |
| Terminal-Bench 4.0 (AA) | 65.2% | 57.1% | 59.1% | 59.6% |
| FrontierSWE | 61.2% | 55.0% | 65.5% | 62.3% |
| OSWorld-2.1 | 81.5% | 74.7% | 80.9% | 87.2% |
| GraphWalks 256k to 1M | 96.2% | 84.2% | 71.8% | 66.8% |
| SEC-bench Pro | 88.2% | 83.4% | 85.4% | Not listed |
| ExploitGym | 54.8% | 42.0% | 42.4% | 34.5% |
| CWE-bench | 71.0% | 68.0% | 68.0% | 67.0% |
The table shows Carbon is not first everywhere. GPT-6 Astra leads on FrontierSWE and Claude Opus 5.5 leads on OSWorld-2.1. Carbon’s largest lead is on long-context retrieval (GraphWalks) and on exploitation (ExploitGym). For a lower-cost OpenAI option, see our GPT-6.1 Terra guide.
Who can use it and the Fairwind Program
Google says it is taking a deliberate approach and is taking part in the U.S. government’s voluntary pre-release model testing. Starting the week of October 28, Carbon goes to a subset of cyber defenders, then to more trusted partners through the Fairwind Program. Developers, businesses and consumers come after that, with no date given.
The Fairwind page sets the rules for partners. Priority goes to governments and national cyber authorities, critical infrastructure operators (such as healthcare, telecommunications, energy and finance), and core technology platforms. Partners must use phishing-resistant multi-factor authentication. They may grant access only to internal cybersecurity, incident response or penetration testing teams, and may not share or resell access. The Fairwind page itself still describes Argon access only.
How to prepare for Gemini 4 Carbon
Carbon is not in the public API yet, but you can prepare now.
- If you work for an eligible security organization, contact your Google Cloud account team to request Fairwind access, as the Fairwind page instructs.
- Build and test your workflow on
gemini-4-argonin Google AI Studio or the Gemini API now. Argon is not on the free tier, so use a paid-tier project. - Store the model name in one configuration value rather than hard-coding it, so you can swap in the Carbon ID once Google publishes it.
- Save a few of your own test tasks and record Argon’s results to compare later.
- Budget with the published prices: $2 input and $10 output per million tokens at launch, then $4 and $20. For Argon, thinking tokens are billed as output.
- Watch the Gemini API release notes for a Carbon entry, a model ID and limits.
The release notes also list two recent API additions you can use with Argon today: the Gemini API CLI (gemini-api) beta on October 27 and Computer Use general availability on October 30. For consumer features, see our Gemini app new features guide.
Workflows that suit Carbon
These prompt patterns follow Google’s description of Carbon. Test them on your own tasks.
- Find, validate, patch. Ask for the three steps separately: list suspected vulnerabilities with file and line, explain how each could be exploited in your environment, then propose a root-cause fix with a test.
- Long debugging sessions. Give the goal, the constraints and a clear stop condition, and ask the model to report dead ends and what it tried.
- Ask for caveats. Ask it to flag missing evidence and discrepancies before it acts.
If you use agentic coding tools, compare this with Anthropic’s approach in our Claude Code guide.
Limitations
- There is no public access, model ID, context window, rate limit or product surface yet.
- Google has not published a Carbon model card. The DeepMind page links an evaluation methodology document.
- All benchmark figures come from Google, and the security results were measured without guardrails, so results under deployed safeguards may differ.
- The end of Carbon’s introductory pricing is not stated.
- Paid-tier Gemini API projects created after October 26 store Generate Content requests by default, according to the release notes. Review that setting before sending sensitive code, and see our answer on whether Gemini trains on your data.
Frequently asked questions
What is Gemini 4 Carbon?
Gemini 4 Carbon is Google’s newest frontier model, announced on October 28, 2026. Google calls it its strongest model for cybersecurity and long-horizon coding and agent work. It builds on Gemini 4 Argon, which launched about two weeks earlier.
Can I use Gemini 4 Carbon now?
Not unless your organization is in Google’s first group. Google says Carbon is going first to a subset of cyber defenders, then to more trusted partners through the Fairwind Program, and to developers, businesses and consumers as soon as possible. Google has not given a date.
How much does Gemini 4 Carbon cost?
Google says Carbon will launch at Argon’s introductory price of $2 per million input tokens and $10 per million output tokens, with cached input 95% off. After the introductory period, the price becomes $4 input and $20 output per million tokens.
What is the Gemini 4 Carbon model ID?
Google has not published one. As of October 31, 2026, the Gemini API changelog and pricing pages list gemini-4-argon as the only Gemini 4 model, and they have no Carbon entry.
Is Gemini 4 Carbon better than Gemini 4 Argon?
On Google’s published table, Carbon scores higher than Argon on every listed benchmark, for example 59 versus 50 on the Artificial Analysis Intelligence Index v5.0 and 54.8% versus 42.0% on ExploitGym. These are Google’s figures, and the security tests were run without guardrails.
Sources
- Google Blog: Gemini 4 Carbon, our next frontier model
- Google DeepMind: Gemini models and benchmark table
- Google DeepMind: Fairwind Program
- Gemini API pricing
- Gemini API release notes
Image: Google Gemini logo by Google LLC, public domain (trademarked), via Wikimedia Commons. Placed on a white background and resized. Gemini is a trademark of Google LLC. FreeAnswers is not affiliated with or endorsed by Google.
1 thought on “Gemini 4 Carbon: Power-User Guide to Access, Pricing, Benchmarks and Limits”