Google DeepMind Releases Gemini 4 Carbon with Multimodal Agentic Reasoning
Frontier AI models · AI assistants · Cybersecurity and privacy
Google DeepMind announced the availability of Gemini 4 Carbon, an advanced AI model engineered for deep reasoning, multi-step agentic workflows, cybersecurity, and scientific tasks. Google DeepMind also previewed the /swarm feature in Google Antigravity.
On November 10, 2026, Google DeepMind announced the general availability of Gemini 4 Carbon for developers, enterprises, scientists, and consumers. Built to test the limits of intelligence in open-ended scientific research, complex coding, and multi-step reasoning, Carbon is designed to run extended verification loops. Google DeepMind also introduced a preview of the /swarm multi-agent orchestration feature within Google Antigravity.
Key takeaways
- Google DeepMind made Gemini 4 Carbon available via the Google AI Studio API and Gemini for business at $2 per million input tokens and $10 per million output tokens, alongside rollouts to Ultra subscribers.
- The model is designed for long-horizon agentic workflows, complex coding, debugging, formal code verification, and scientific analysis, utilizing more tokens than Gemini 4 Argon for verification loops.
- Google Antigravity previewed /swarm, enabling users to coordinate multi-agent teams from a single prompt, with demonstrations reconstructing 3D environments from planning drawings and historical photos.
- According to Google DeepMind, the model demonstrated cybersecurity defenses by identifying and validating zero-day software vulnerabilities while incorporating safeguards against prompt injection and misuse.
Source：Google DeepMind Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human.