Google is about to roll out a new AI model. Employees say they're testing another that's way better.

Alphabet's Google is in the process of testing a new AI model that may outperform its flagship Gemini 4 system, according to employees familiar with the development. The latest internal release, codenamed Carbon, has been compared by some workers to Anthropic's Opus 5.5 model, which is considered one of the most powerful models for long-term agentic coding.

The testing of Carbon, which may be an update to Google's Argon model, has been documented via internal Google platforms and screenshots obtained by Business Insider. One employee described Carbon as "feeling like Opus 5.5 for coding," hinting at its superior performance. However, others noted that early versions of Argon, Google's previous attempt at an AI coding agent, fell short in certain tasks, suggesting that Google is racing against rivals like Anthropic and OpenAI to develop advanced AI coding agents.

Google's internal tests aim to close the gap with rivals, but there is no guarantee that the improved versions will be released publicly. The company's internal codenames for these models, such as Argon, Barium, and Carbon, do not align with their public names, leading to speculation about whether Carbon is an upgraded Argon model or a separate update within the Gemini 4 family.