Gemini 4 Carbon Frontier Workspace
Built to stay on task for days and to defend systems at the highest level
None

None

None

Long Story Video Skill

Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.

Ads Video Skill

Ads Video Skill

Generate professional ads and sales videos—AI auto-generates scripts, prompts, and footage.

3D Science Explainer Video Skill

Convert scientific concepts into stunning 3D explain animations

AI Video Prompt Generator

Feedback

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

  • ✓freeTrialImage.benefitHd
  • ✓freeTrialImage.benefitWatermark
  • ✓freeTrialImage.benefitUnlimited

Gemini 4 Carbon

Gemini 4 Carbon keeps multi-day coding agents on task and hardens cyber defense. Compare its benchmark scores, token pricing and rollout here.

All Tools

Discover our comprehensive AI-powered animation toolkit

Where Gemini 4 Carbon Excels — and Why It Matters

This release is tuned for work that spans days rather than minutes and treats security as a first-class job, all behind a deliberately careful rollout.

  • Agents That Keep Going for Days
    The model was tuned for engineering problems that resist one-shot answers. Early internal runs lasted days at a stretch, pushing past dead ends to trace and repair subtle faults in huge codebases and low-level system code.
  • Security Work from Detection to Fix
    Coverage runs the whole arc — spotting real weaknesses at 88.2% on SEC-bench Pro, attempting exploitation at 54.8% on ExploitGym, and closing the root cause at 71.0% on CWE-bench.
  • Two Weeks from Argon to Carbon
    Arriving just a fortnight after Gemini 4 Argon, this release shows how rapid iteration between researchers and models moves the frontier forward while safeguards are hardened alongside.

The Staged Release of Gemini 4 Carbon

Availability is opening in careful stages rather than all at once — this is the order in which it happens.

What Gemini 4 Carbon Delivers, by the Numbers

A first-place Intelligence Index score, a 0.0% prompt-injection success rate and a near-80% drop in misalignment alerts — here is the evidence.

First Place on Intelligence Index v5.0

A score of 59 puts it at the top of the freshly revised Artificial Analysis Intelligence Index v5.0, with the biggest jumps landing in coding, agentic execution and the security skills that matter most.

Climbing to #4 on WebDev Arena

Cleaner output and better artifacts in full-stack builds add +89 elo points, moving it to fourth place at 1768 on WebDev Arena.

Shortcut-Taking Cut to Zero

Alignment work brought successful cheating on Terminal-Bench 4.0 down to 0.0% and attempts to just 0.4% while accuracy rose, and high-severity misalignment alerts dropped almost 80% across internal live coding environments.

Straight Answers, Fewer Overclaims

Internal testing shows unsupported claims falling by roughly half versus Argon; the model now writes methodological caveats into the deliverable or declines blocked actions with a clear list of what would still be needed.

Zero Traction for Indirect Prompt Injection

On Gray Swan's Indirect Prompt Injection Benchmark every evaluated attack failed — a 0.0% success rate, down from the 0.7% Argon recorded.

Finishing the Job, Staying in Scope

Overreach dropped 26% and under-service fell 29%, so multi-step plans now run end to end more reliably while respecting project constraints and staying strictly in scope.

FAQ

Gemini 4 Carbon: Questions Answered

Token costs, headline benchmark scores, early access and how this release differs from Argon — all answered below.

1

What exactly is Gemini 4 Carbon?

The newest and most capable frontier release from the team, aimed at state-of-the-art security work and long-horizon tasks. It takes first place on the updated Artificial Analysis Intelligence Index v5.0 with a score of 59.

2

What does it cost per token?

Launch pricing is $2 per million input tokens and $10 per million output tokens, with cached input at 95% off the input rate. Standard rates of $4 per 1M input and $20 per 1M output begin after the introductory window.

3

Where does Carbon lead on benchmarks?

It hits 88.2% on SEC-bench Pro for spotting vulnerabilities, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching, and lands at #4 on WebDev Arena with 1768 points.

4

Who gets access today?

A hand-picked cohort of cyber defenders is first; more trusted partners join over time under the Fairwind Program, followed by developers, businesses and consumers through Google's product surfaces.

5

How does it compare with Gemini 4 Argon?

Internal benchmarks show more than 2x uplift in generating secure code, roughly half as many unsupported claims, a 0.0% prompt-injection attack success rate and tighter scope discipline.

6

What work is it designed for?

Hard, multi-step engineering — root-causing and fixing subtle faults in large codebases and low-level systems — along with agentic workflows, scientific tasks, full-stack web development and multimodal understanding.

Gemini 4 Carbon Is Coming — Be First in Line

The goal is a real step change in software engineering and cyber defense, with safety and reliability kept intact. Track the staged rollout through the Fairwind Program as developer, business and consumer access opens.