Google releases Gemini 4 Carbon AI model with 1M-token context window
Also covered by 9 sources
What happened
Google has launched Gemini 4 Carbon, its most advanced AI model, making it available globally to developers, enterprises, scientists, and AI Ultra subscribers. The model, initially released to select cybersecurity professionals, is now accessible through Google AI Studio, the Gemini app, and AI Gateway. Gemini 4 Carbon is designed for complex tasks such as multi-step coding, advanced cybersecurity, and scientific research. It introduces a new agent 'swarm' feature and offers a 1 million token context window for handling large and intricate projects.
Key details
- Gemini 4 Carbon features a 1 million token context window.
- The model is available at launch for $2 per million input tokens and $10 per million output tokens, with a 50% discount on AI Gateway until November 30, 2026.
- A new 'swarm' capability enables coordinated teams of AI agents for complex workflows.
- 10,000 scientists have been granted free six-month access for research and model training.
- Gemini 4 Carbon outperformed competitors on benchmarks like CursorBench 4.0 and AutomationBench, and demonstrated strong resistance to prompt-injection attacks.
Why it matters
Developers, enterprises, and researchers now have access to a more powerful tool for complex coding, cybersecurity, and scientific analysis, enabling them to tackle larger and more intricate projects. The expanded context window and agent collaboration features allow for deeper reasoning and longer workflows, which can reduce manual effort and improve productivity. The competitive pricing and research access program lower the barrier for organizations and scientists to adopt advanced AI capabilities in their work.