Gemini 4 Carbon
Gemini 4 Carbon is Google's flagship reasoning model, engineered specifically for rigorous software development, architectural refactoring, and multi-hour engineering workloads.
- Gemini 4 models
Compare the four models, Standard API prices and personal Gemini app access.
- Gemini 4 Neon API reference
Check the request model ID, Standard rates, thinking levels and output limits.
Browse Language Models models to compare other options for this task.
🚀Function Overview
Google announced gemini-4-carbon on October 28, 2026, and launched it to Developer API general availability as a paid model on November 10, 2026. It processes text, images, audio, video, and PDFs across a 1,048,576-token context window, outputting text up to 65,536 tokens synchronously or up to 1,048,576 tokens via optional long decode. Carbon supports low, medium, and high thinking modes with medium set as default. Developer API pricing is $2 per million input tokens and $10 per million output tokens through November 30, 2026, transitioning to $4 per million input tokens and $20 per million output tokens on December 1, 2026.
Key Features
- Google’s most capable reasoning model for complex software engineering and long-horizon tasks.
- Unified 1,048,576-token context window accepting text, image, audio, video, and PDF inputs
- Carbon supports low, medium, and high thinking levels. Medium is the Developer API default.
- In the Interactions API, one non-streaming hop on Carbon or Argon is capped at 32,768 tokens. Streaming hops and Neon or Hydrogen non-streaming hops can use up to 65,536. Long decode can continue within the shared 1,048,576-token context.
Use Cases
- •Refactoring complex enterprise software systems and resolving cross-repository dependencies
- •Work through multi-step debugging tasks and verify the fix with regression tests.
- •Analyzing extensive code repositories alongside technical audiovisual specifications and system logs
⚙️Input Parameters
contents
arrayGemini generateContent contents with text or supported media parts
generationConfig
objectDocumented output and thinking configuration; ordinary synchronous output is distinct from long decode
💡Usage Examples
Example 1
Input Parameters
{
"contents": [
{
"role": "user",
"parts": [
{
"text": "Review the supplied project notes. List confirmed findings, open questions, and the checks that would resolve them."
}
]
}
]
}Output Results
Quick Actions
Technical Specifications
- Hardware Type
- Hosted Google API
- Commercial Use
- Hosted API use follows Google’s terms.
- Pricing
- Gemini Developer API Standard. $2 input, $10 output and $0.10 cached input per million tokens through November 30, 2026. From December 1: $4 input, $20 output and $0.20 cached input.