Google Releases Gemini 4 Carbon Frontier Language Model

Carbon also shows massive improvements in coding intelligence, jumping +89 Elo to reach a score of 1769 on the WebDev Arena. In terms of safety and adversarial robustness, the model achieved a 0.0% attack success rate on Gray Swan's Indirect Prompt Injection benchmark, a significant improvement over Argon's 0.7%. Furthermore, it reduced successful cheating on Terminal-Bench 4.0 to 0.0%, proving its integrity in controlled execution environments.

• SEC-bench Pro (Vulnerability finding): 88.2%
• CWE-bench (Patching): 71.0%
• ExploitGym (Exploitation): 54.8%
• WebDev Arena: 1769 Elo (#4)
• Indirect Prompt Injection: 0.0% attack success rate

API Pricing

Google has maintained a competitive pricing strategy for Carbon, matching the introductory rates of the Argon model to ensure a smooth transition for developers. This makes high-end cybersecurity and engineering capabilities accessible for both startups and large-scale enterprises.

A standout feature is the aggressive pricing for cached inputs. By offering a 95% discount on cached tokens, Google is incentivizing developers to build RAG-heavy applications and long-context engineering workflows where the same codebase or documentation is frequently referenced.

• Input Tokens: $2.00 per million
• Output Tokens: $10.00 per million
• Cached Input: $0.10 per million (95% discount)

Use Cases

Gemini 4 Carbon is not a general-purpose chatbot; it is a specialized tool for technical professionals. Its primary use case is in Automated Security Operations (SecOps), where it can identify, analyze, and patch vulnerabilities in real-time. It is also uniquely suited for 'Deep Engineering'—tasks that require a model to understand a massive repository, trace a bug through multiple layers of abstraction, and propose a verified fix.

Developers can also leverage Carbon for high-fidelity agentic workflows. Because it has significantly reduced 'under-service' and 'overreach' in multi-step planning, it is ideal for building autonomous DevOps agents that manage CI/CD pipelines, infrastructure as code (IaC), and complex system migrations.

• Cybersecurity: Vulnerability research and automated patching
• Software Engineering: Long-context debugging and root-cause analysis
• DevOps: Autonomous multi-step infrastructure management
• Agentic Workflows: High-precision planning and execution

Getting Started

Access to Gemini 4 Carbon is being managed through a staged rollout to ensure safety and stability. Initially, the model is available to a select group of cyber defenders and security researchers. Following this phase, access will expand to trusted partners through the Fairwind Program.

General availability for developers, businesses, and consumers will follow shortly after the Fairwind phase. Developers can prepare by monitoring the Google AI Studio and Vertex AI platforms for official API endpoint availability and SDK updates.

• Phase 1: Selected Cyber Defenders
• Phase 2: Fairwind Program (Trusted Partners)
• Phase 3: General Developer and Business Access

API Pricing — Input: $2.00 per million tokens (introductory) / Output: $10.00 per million tokens (introductory) / Context: Cached input available at $0.10 per million tokens (95% off)

Sources