Google is internally testing a new Gemini 4 model codenamed Carbon, aiming to improve its coding capabilities and compete more effectively with rivals like Anthropic and OpenAI.
Key Points
- Google employees are testing an internal model called Carbon, which some staff compare to Anthropic’s high-performing Claude Opus 5.5 for coding tasks.
- Carbon is part of the broader Gemini 4 family, which includes other internal iterations such as Argon and Barium.
- The upcoming Gemini 4 "Argon" model is designed for frontier performance in legal, finance, and defensive cybersecurity sectors.
- Google plans to release Argon to partners in its Fairwind Program for vulnerability testing before a wider public rollout.
- The company is accelerating development to maintain its competitive standing in the rapidly evolving market for AI-powered coding agents.