IT Home On October 10, according to Business Insider, Google is about to officially release the new generation Gemini 4 “Argon” model to the public. Meanwhile, its internal staff are testing a more powerful subsequent iteration version, which is expected to further narrow the technical gap with competitors.
According to internal documents and screenshots obtained by this media, Google employees have been testing a new Gemini 4 version codenamed “Carbon” recently. This model is likely a new checkpoint update under the Argon architecture, and it has been integrated into Google’s internal code platform Jetski in recent days.
An employee revealed that Carbon’s performance in code capabilities is “comparable to Opus 5.5” – the strongest model created by Anthropic for long-term tasks involving complex code agents. However, the employee also said that this conclusion needs further in-depth testing and validation.
Based on feedback from multiple employees and internal communication records, the early version of Argon received positive reviews in internal tests. However, some employees noted that the early version was still challenging for certain programming tasks, and its level was roughly equivalent to Claude Opus 5 released by Anthropic earlier.
An internal document indicates that Google has tested several Gemini 4 derivative models codenamed Argon, Barium, and Carbon.
It is worth noting that Google’s internal product codenames do not always correspond with the final public names. For example, according to documents, the model that was originally chosen for public release under the name “Argon” actually had an internal codename of “Barium-B”. Therefore, it is still unclear whether the current Carbon will be a refined version of Argon or a new independent model under the Gemini 4 family.
Regarding this, Google refused to comment.
Related Reading: