Google is currently testing a new version of its Gemini 4 AI model, internally referred to as Carbon. The model is being evaluated by company employees on the internal Jetski platform, which is primarily used for code generation tasks.
Internal feedback suggests that the Carbon version performs similarly to the Claude Opus 5.5 model from Anthropic in programming tasks. Employees noted that while initial results are promising, more extensive testing is required to fully assess the model's capabilities.
The development of Gemini 4 involves several internal iterations, including versions codenamed Argon, Barium, and Carbon. Google previously confirmed that the Barium-B iteration will be released publicly under the name Argon.
Google aims to strengthen its position in the competitive AI agent market, where Anthropic and OpenAI currently hold significant leads. The company is focusing on enhancing its models' performance in complex engineering, legal, and financial tasks.
The upcoming Argon model will initially be available to partners through the Fairwind program, which focuses on identifying security vulnerabilities. A wider public release is planned for a later date, following the initial testing phase.
