Google is preparing the public release of its next-generation Gemini 4 model, internally referred to as Argon. According to internal documents, the company is also testing a more powerful iteration codenamed Carbon, which is currently integrated into the internal Jetski code platform.
Early internal feedback suggests that the initial version of Argon performs at a level comparable to Anthropic's Claude Opus 5 in programming tasks. Some employees noted that the model still faces challenges with certain complex coding assignments, though testing remains ongoing.
The Carbon variant is reportedly showing performance in code-related tasks that rivals Anthropic's Opus 5.5, a model designed for long-term complex agentic tasks. However, staff members emphasize that these performance claims require further rigorous evaluation.
Google is currently testing several derivatives of the Gemini 4 architecture, including models codenamed Argon, Barium, and Carbon. It remains unclear whether Carbon will be released as a fine-tuned version of Argon or as a standalone model within the Gemini 4 family.
The company frequently uses internal codenames that differ from final product names. For instance, the model being released as Argon was internally designated as Barium-B. Consequently, the final branding for the current testing projects has not yet been determined.
