Gemini 4 Carbon
by Google DeepMind
Google's frontier reasoning model for cybersecurity and coding.
Coming soon.
Use Gemini 4 Argon on Morphic today.

Key features
Features and specs reflect what Google has published so far. This page updates with the launch specs.
Technical specifications
Reasoning
The frontier tier of the Gemini 4 family; writes text, not media
5 types
Text, images, video, audio and PDF (same Gemini 4 family)
Text
Code, plans, prompts and tool calls
1M tokens
Same family window as Gemini 4 Argon
#1 AI Index
Score of 59 on AI Intelligence Index v5.0
$2 / $10 per 1M
Introductory rate per million tokens via API
Fairwind
Cyber defenders first, broader access coming
0% injection
Zero prompt-injection attack success rate
Use cases
Vulnerability assessment
Analyze a codebase or system configuration for security weaknesses and get a ranked list of findings with remediation steps.
Secure code generation
Write code that follows security best practices from the start, with more than 2x the secure-code success rate of its predecessor.
Agentic security workflows
Chain together scanning, analysis and patching steps that run across tools and repositories without losing track of context.
Code review and hardening
Review pull requests for security issues, suggest hardened alternatives and flag patterns that commonly lead to exploits.
Threat modeling
Map an application or infrastructure against known attack surfaces, prioritize risks and plan defenses.
Long-horizon planning
Break large multi-step projects into ordered tasks, track dependencies and carry context across a long chain of decisions.
Prompt examples
Security review
Review this codebase for vulnerabilities and rank the findings by severity
Edit promptIncident plan
Draft an incident response plan for a credential exposure in this infrastructure
Edit promptGemini 4 Carbon overview
Gemini 4 Carbon is Google's frontier reasoning model of the Gemini 4 generation. It is designed for long-horizon agentic work, coding and cybersecurity, and it scored 59 on the AI Intelligence Index v5.0, the highest of any model at the time of announcement. The Fairwind Program has opened early access to cyber defenders, with broader access in the coming weeks.
Gemini 4 Carbon release date and specs
Google has not published a general availability date. The Fairwind Program has opened early access to cyber defenders, and broader access is expected in the coming weeks. API pricing starts at $2 per million input tokens and $10 per million output tokens during the introductory period, moving to $4 and $20 after.
Gemini 4 Carbon and Morphic
Carbon is not on Morphic yet. Its predecessor, Gemini 4 Argon, runs in Morphic Copilot on every paid plan from Basic up, and it shares Carbon's reasoning approach and multimodal input. Use Argon for the same kind of multi-step planning and production work while Carbon access widens. The Gemini 4 Argon guide walks through Copilot and MCP setup.
The Gemini line runs on Morphic. Likely, because of that pattern, the platform would add Carbon once broader access opens.
Limits to plan around
Carbon is a reasoning model: it writes text, not media. On a platform that provides image, video and audio models, it would plan and direct while those models render. Benchmark figures are measured under controlled conditions; results on production tasks depend on the task and the data. Access is currently limited to cyber defenders, and the introductory API pricing has a standard rate behind it.
Simple pricing
Get started for free today, with the option to upgrade or cancel anytime.
Basic
2400 monthly credits
1 user only
All models
Workflows
Standard
3625 monthly credits
1 user only
All models
Workflows
Pro
6350 shared monthly credits
1 user
All models
Workflows
Max
24650 shared monthly credits
1 user
All models
Workflows
Enterprise
For higher limits
Custom
pricing and billing terms

Free
For playing around
$0
forever free
FAQs
Seedream 5.5
ByteDance
ByteDance's Seedream image model, released November 2026. Clean multi-round edits, lifelike realism, and sharper text.
Gemini Omni Flash 1.2
Google DeepMind
Google's video model, now fifteen seconds long. Edit the whole take for 30% less a second.
ChatGPT Images 2.6
OpenAI
OpenAI's image model, released November 2026. Animated GIFs, SVG export, faster drafts.
SAM 3.1
Meta
Meta's Segment Anything model. Track any subject as a mask that follows it.