AI News technology

Gemini 4 Carbon Public Release Date: What Google Has Announced

Gemini 4 Carbon launched broadly on November 10, 2026, with paid API access and a Gemini app rollout for Google AI Ultra subscribers.

Gemini 4 Carbon launched broadly on November 10, 2026, with paid API access and a Gemini app rollout for Google AI Ultra subscribers.

0 Comments
November 10, 2026Public release
Paid tierAPI access
Google AI UltraGemini app
$2 introductoryInput / 1M
$10 introductoryOutput / 1M
November 30, 2026Offer ends

The short version

  • Gemini 4 Carbon became generally available on November 10, 2026.
  • Paying developers and enterprises can access its API.
  • Google AI Ultra subscribers are getting Carbon in Google’s consumer apps.
  • Introductory API pricing ends on November 30, 2026.
  • From December 1, 2026, input costs $4 and output costs $20 per million tokens.

Is Gemini 4 Carbon available to everyone now?

Gemini 4 Carbon is no longer restricted to selected cyber defenders: since November 10, 2026, paying developers and enterprises can access it through Google AI Studio and Gemini for business, while Google AI Ultra subscribers are getting it in the Gemini app, AI Mode in Search and Google Antigravity. Google’s November 10, 2026 availability announcement sets introductory API pricing at $2 for a million input tokens and $10 for a million output tokens through November 30, 2026, rising to $4 and $20 respectively from December 1, 2026. Google initially announced Carbon on October 28, 2026, with access beginning among selected cybersecurity defenders.

General availability comes with account requirements. Developers need the Gemini API paid tier, and Google’s announced consumer rollout requires Google AI Ultra.

Gemini 4 Carbon release date for developers, businesses and consumers

November 10, 2026 is Carbon’s general release date for paying developers and enterprises, and the start of its rollout to Google AI Ultra subscribers.

Access dates and supported products

Date Who got access Where
October 28, 2026 Initial rollout announced for selected cyber defenders Fairwind Program
November 10, 2026 All paying developers Gemini API paid tier in Google AI Studio and Android Studio
November 10, 2026 Enterprises Gemini for business through its API or model selector, Google Antigravity, CodeMender
November 10, 2026 Google AI Ultra subscribers Gemini app, AI Mode in Search, Google Antigravity, Google AI Studio

Google describes the Google AI Ultra release as a rollout. API access for developers follows the paid-tier requirement.

In its October 28, 2026 announcement, Google described a staged rollout and said it was participating in voluntary U.S. government testing before release.

What is Gemini 4 Carbon used for?

Google describes Carbon as its strongest model, intended for difficult tasks that need sustained reasoning and repeated checks.

Google identifies these areas of work:

  • Software engineering, including debugging and changes across large codebases.
  • Cybersecurity, including finding vulnerabilities and producing patches.
  • Knowledge work, including analysis of documents and structured deliverables.
  • Science, including research questions and reasoning through experimental data.

Google says its internal teams observed Carbon working for days on engineering problems, trying other approaches after unsuccessful attempts. Those tasks included tracing elusive bugs and developing fixes across extensive codebases and low-level systems.

Google says Carbon can spend more tokens than Argon investigating problems and checking its work. Google still recommends Argon for most software engineering and knowledge work.

Google is also previewing /swarm in Google Antigravity. It coordinates agents from one prompt, with Carbon handling demanding reasoning roles while other models work in separate workspaces and verifier agents check changes.

What is the Gemini 4 Carbon price?

Carbon’s introductory API rate is $2 per million input tokens and $10 per million output tokens through November 30, 2026.

API token rates

Item Through November 30, 2026 From December 1, 2026
Input per 1M tokens $2 $4
Output per 1M tokens $10 $20

Google specifies these rates for the publicly accessible Carbon model. Its introductory input and output prices match Argon’s.

Google applies a 95% discount to the input-token rate for cached tokens.

What benchmark results has Google reported for Carbon?

Google reports results across coding, cybersecurity, knowledge work and safety evaluations. Google specifies that the cybersecurity capability scores published on October 28, 2026 were evaluated without guardrails.

Google’s reported evaluation results

Evaluation Google’s reported Carbon result
Artificial Analysis Intelligence Index v5.0 Google: 59, ranked #1
SEC-bench Pro Google: 88.2% for finding vulnerabilities
ExploitGym Google: 54.8% for vulnerability exploitation
Collinear’s CWE-bench Google: 71.0% for patching vulnerabilities
WebDev Arena Google: 1768, ranked #4, with a gain of 89 Elo points
CursorBench 4.0 Google: ranked #1
WANDR Google: ranked #1
AutomationBench Google: ranked #1
Multimodal Knowledge Agent benchmark Google: ranked #2, with average task cost more than 6X lower than the top-performing model
General Financial Intelligence benchmark Google: top placement among 23 frontier model configurations in pre-release evaluations
Project Fortify internal benchmark Google: 100% scanning recall and 95% triage precision
Open-source repository testing Google: recall improved by over 25% relative to Argon, uncovering previously missed logic flaws in real software
Internal secure-code generation benchmark Google: more than 2x Argon’s success rate
Chrome team’s patch-generation comparison Google: Carbon’s performance on producing quality patches exceeded another lab’s large frontier model by more than 10X
Internal coding deployments Google: alerts for high-severity misalignment fell by almost 80%
Internal unsupported-claim testing Google: approximately 2x fewer unsupported claims than Argon
Multi-step task execution Google: overreach fell by 26%, while under-service fell by 29%
Terminal-Bench 4.0 cheating evaluation Google: successful cheating at 0.0%, with attempts at 0.4%
Agentic honesty benchmark Google: 92% of 375 tasks passed, compared with 70% for Argon; missing experimental data was left unfilled 98% of the time, and methodological problems were flagged 68% of the time
Adaptive simulated cyberattack evaluation Google: 99% of attack trajectories blocked; per-attempt breach rate of 0.8%, compared with 3.1% for Argon
Gray Swan automated cyberattack testing Google: query success rate of 2.17%, compared with 9.86% for Argon
Gray Swan Indirect Prompt Injection benchmark Google: attack success rate of 0.0%, compared with 0.7% for Argon
Trajectory Labs external testing Google: zero new universal jailbreaks across 132 hours and more than 13,000 requests

Google also reports that Carbon matched the best model evaluated in Hebbia’s E2E Financial Agentic evaluation. For CursorBench 4.0, WANDR and AutomationBench, Google reports task costs more than 2X lower than the next-best external model.

Google says defenses against prompt injection still need further improvement.

Google also reports that Wiz used Carbon to find zero-day vulnerabilities while examining two major SaaS providers, taking less than two hours for work it described as weeks of vulnerability research.

What we saw in the Gemini app and the API

On November 12, 2026, we read the Gemini app’s menus in desktop Chrome on Windows using our Google AI Pro account, and read the model list returned for our Gemini API key.

The app displayed Most complex work with 4 Carbon underneath and a lock indicating it was unavailable on that Pro account. The same menu offered Everyday help with 4 Hydrogen and Code & analysis with 4 Argon.

Our API model list included gemini-4-carbon and showed an output token limit of 32,768 for that entry. These observations came from reading the menus and model list; we sent no prompts and evaluated no answers.

Gemini 4 Carbon vs Gemini 4 Argon: which should I use?

Google recommends Argon for most software engineering and knowledge work, while positioning Carbon for harder problems that require extended investigation and verification.

Google’s stated model roles

Detail Carbon Argon Hydrogen
Google’s role for the model Demanding engineering, cybersecurity and scientific reasoning Most software engineering and knowledge work Everyday developer coding and agent workflows

For Gemini 4 Carbon vs Argon, Google’s distinction is the work involved: Carbon is intended to explore difficult problems more extensively, while Argon remains Google’s recommendation for the majority of coding and knowledge tasks.

Do defenders and scientists still have separate access programs?

Google continues to offer Carbon for cyber through the Fairwind Program, with defenders able to apply for admission on a rolling basis.

Google’s scientists’ program also includes Carbon for existing and future participants. Selected researchers can receive six months of free access to Argon and Carbon through Google Antigravity’s highest usage tier. Applications remain open, with weekly onboarding governed by eligibility, selection and capacity.

In its November 10, 2026 announcement, Google said applications had opened four weeks earlier for a program offering 10,000 scientists six months of Argon access in Google Antigravity at no charge.

FAQs

What is the Gemini 4 Carbon public release date?

Gemini 4 Carbon became generally available on November 10, 2026. Paying developers and enterprises received access, and Google began rolling it out to Google AI Ultra subscribers.

Can I use Gemini 4 Carbon in the Gemini app with Google AI Pro?

Google’s announced Gemini app rollout is for Google AI Ultra subscribers. Google AI Pro is not the plan named for that access.

Is Gemini 4 Carbon free?

Its standard developer access requires the Gemini API paid tier, and its announced consumer rollout requires Google AI Ultra. Selected scientists can receive free access through Google’s research program.

Can scientists get Gemini 4 Carbon for free?

Yes. Scientists and researchers can apply for six months of free Argon and Carbon access through Google Antigravity’s highest usage tier. Admission depends on eligibility, selection and capacity, with onboarding taking place weekly.

How much will Gemini 4 Carbon cost after November 30, 2026?

From December 1, 2026, the public Carbon model’s API rate will be $4 per million input tokens and $20 per million output tokens.

Is Gemini 4 Carbon better than Gemini 4 Argon?

Google describes Carbon as its strongest model and positions it for especially demanding reasoning tasks. Google still recommends Argon for most software engineering and knowledge work.

Do I still need the Fairwind Program to use Gemini 4 Carbon?

No. General Carbon access is available to paying developers and enterprises, with a consumer rollout for Google AI Ultra subscribers. The Fairwind Program remains the application route for Carbon for cyber.

Leave a Reply

Your email address will not be published. Required fields are marked *