🔥 Black Friday AI API deals are live· New drops every MondayEnds in--d --:--:--See deals →

Google Gemini 4 Carbon: AI Large Language Models

Run Google Gemini 4 Carbon via API: security-focused frontier reasoning for code audit and long-horizon coding. Try free — pay per generation, no subscription.

Interactive model controls

Gemini 4 Carbon is Google DeepMind's security-focused frontier model, ranked first on the Artificial Analysis Intelligence Index and strong at vulnerability discovery, patching, and secure code generation. Accepts text and image input with an optional system prompt and reasoning effort. Token-based pricing: $4.00/M input tokens and $20.00/M output tokens.

📝

Overview

About this model

Gemini 4 Carbon is Google DeepMind's security-focused frontier model. It ranks first on the Artificial Analysis Intelligence Index and posts state-of-the-art results on vulnerability discovery (SEC-bench Pro), exploitation (ExploitGym), and patching (CWE-bench), while also improving secure code generation over Gemini 4 Argon. It accepts text plus an optional image and system prompt, with a reasoning_effort control (low, medium, high). For the general-purpose Gemini 4 tier, see Gemini 4 Argon; for faster, cheaper chat, see Gemini 3.8 Flash.

1Secure Code Review: Audit functions and pull requests for injection, memory-safety, and authentication flaws, then draft patches.
2Vulnerability Triage: Classify findings by CWE, explain exploitability, and prioritize remediation.
3Long-Horizon Coding: Plan and implement multi-step engineering work with strong reasoning.
4Agent Reasoning: Use as the planner model inside multi-step agents and [workflows](/workflows).
5Visual Q&A: Pass an image_url to ask questions about screenshots, architecture diagrams, or scanned documents.
💰

Pricing & Value

Cost analysis

muapiapp$4.00/M input, $20.00/M output

Token-based billing. Minimum $0.0004 per call.

Fal.aiNot available

No dedicated Gemini 4 Carbon endpoint at this time.

ReplicateNot available

No dedicated Gemini 4 Carbon endpoint at this time.

** Competitor pricing is estimated based on similar model architectures and usage tiers.

⚙️

Technical Details

Configuration schema

Promptstring

The user message or instruction for the model.

Default ValueAudit this function for memory-safety and injection vulnerabilities, then propose a patch.
Image URLstring

Optional image URL to include as multimodal input.

Default Valueundefined
System Promptstring

Optional system-level instruction to guide model behavior.

Default ValueYou are a helpful assistant that responds concisely.
Reasoning EffortEnum (3 options)

How much reasoning the model applies before answering. Higher effort is slower and uses more output tokens.

Default Valuemedium
📖

Implementation Guide

Developer documentation

How to Use Gemini 4 Carbon

  1. Write your prompt: Put the task in prompt. Optionally add system_prompt to set role, tone, or output format.

  2. Choose reasoning effort: Set reasoning_effort to low, medium (default), or high. Higher effort reasons longer and bills more output tokens.

  3. Add an image (optional): Provide image_url to include an image in the request.

  4. Submit and fetch the result: POST returns a request_id; poll /api/v1/predictions/{request_id}/result or pass a webhook URL.

cURL Example

curl -X POST "https://api.muapi.ai/api/v1/gemini-4-carbon" \
  -H "x-api-key: $MUAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Audit this function for injection vulnerabilities and propose a patch.",
    "system_prompt": "You are a senior application security engineer. Be concise and specific.",
    "reasoning_effort": "high"
  }'
❓

Common Questions

Frequently asked

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google DeepMind's security-focused frontier model, ranked first on the Artificial Analysis Intelligence Index and strong at finding, exploiting, and patching vulnerabilities as well as general coding and reasoning.

How is Gemini 4 Carbon priced?

Pricing is token-based at $4.00 per million input tokens and $20.00 per million output tokens, with a minimum charge of $0.0004 per call. Reasoning tokens count as output.

Does Gemini 4 Carbon accept images?

Yes. Pass an image_url alongside your prompt to ask questions about a screenshot, diagram, or photographed document.

What does reasoning_effort do?

It sets how much the model reasons before answering: low, medium (default), or high. Higher effort improves depth on hard problems but is slower and uses more output tokens.

How do I get the result?

The endpoint is asynchronous: you receive a request_id immediately, then poll the predictions result endpoint or supply a webhook URL to be notified when the text is ready.

How does Gemini 4 Carbon differ from Gemini 4 Argon?

Carbon is the newer, security-focused model with stronger vulnerability-finding and secure code generation results; Argon is the general frontier reasoning tier. Both use the same prompt, image_url, and system_prompt fields, and Carbon adds reasoning_effort.