Robots Atlas>ROBOTS ATLAS
Gemini 4 Carbon

Gemini 4 Carbon

4 Carbon · Family: Gemini
Frontier model in Google's (Google DeepMind) Gemini 4 line. Multimodal (text, image, audio, video, PDF), 1M-token context window, reasoning mode and agentic capabilities.
✓ Active✓ Public accessLLMMultimodalReasoning modelTool-using model📁 Gemini
Context window
1M
tokens
Max output
1,000,000
tokens
Access:APIHostedDeployment:☁ Cloud

Overview

Gemini 4 Carbon is a frontier model in Google's (Google DeepMind) Gemini 4 line, released in November 2026 under the API identifier gemini-4-carbon. It belongs to the same generation as Gemini 4 Argon and Gemini 4 Hydrogen.

The model is multimodal - it accepts text, images, audio, video and PDF files - and offers a 1 million token context window. It supports a reasoning (thinking) mode and agentic tasks, including agentic coding and tool use. It is available through the API (paid tier) with public pricing.

On Google DeepMind's official benchmarks, Carbon scores, among others, 61.2% on FrontierSWE 2, 65.2% on Terminal-Bench 4.0 and 96.2% F1 on GraphWalks at up to 1M-token context.

Classification
LLMMultimodalReasoning modelTool-using model
Family: Gemini
Access & deployment
APIHosted
Cloud
Weights: Closed
Key parameters
📏 Context: 1M
✓ Tools
📥 Input: text, image, audio, video…

Technical specification

Context window
1M
tokens
Max output tokens
1,000,000
tokens per response
License
Proprietary (closed weights)
Features:✓ Tool use
Modalities
⬇ Input
textimageaudiovideodocuments
⬆ Output
textcode

Capabilities and applications

Native model capabilities
Agentic capability
The model's ability to autonomously plan and execute multi-step tasks by sequentially using tools, maintaining context, and adapting to intermediate results.
Category: planning
Agentic coding
Multi-hour, multi-step programming tasks performed autonomously by the model: cloning a repository, running tests, iterating on fixes, integrating with CLI tools. Characteristic of Codex variants (GPT-5.1-Codex-Mini, Codex-Max).
Category: coding
Multimodal understanding
Category: multimodal

Benchmark results

7 benchmarks
AA Index v5.0
59
📄 Google DeepMind (deepmind.google/models/gemini)
Vals Index v2.2
70.4%
📄 Google DeepMind (deepmind.google/models/gemini)
FrontierSWE 2
61.2%
📄 Google DeepMind (deepmind.google/models/gemini)
Terminal-Bench 4.0 (AA)
65.2%
📄 Google DeepMind (deepmind.google/models/gemini)
Agents' Last Exam v1.1
77.4%
📄 Google DeepMind (deepmind.google/models/gemini)
GraphWalks (256k-1M, BFS F1)
F1
96.2%
📄 Google DeepMind (deepmind.google/models/gemini)
SEC-bench Pro
88.2%
📄 Google DeepMind (deepmind.google/models/gemini)

Pricing

Technical architecture