- Published on
Google Gemini 4 'Argon' Model Coming Soon, New 'Carbon' Version Reportedly Being Tested Internally
- Authors
- Name
- DeepHorus
According to Business Insider, Google is about to officially release its next-generation Gemini 4 'Argon' model to the public. At the same time, employees inside the company have already been testing a more powerful follow-up iteration, hoping to further narrow the technological gap with competitors. Google declined to comment on the matter.
The report said that internal documents and screenshots obtained by the outlet show that Google employees have recently been testing a new Gemini 4 version codenamed 'Carbon.' The model is likely a new checkpoint update under the Argon architecture and has recently been connected to Google's internal code platform Jetski. Feedback from multiple employees and internal communication records also indicate that Google has internally begun testing several Gemini 4 derivative models, including Argon, Barium, and Carbon.
In terms of coding ability, one employee revealed that Carbon's performance is 'comparable to Opus 5.5.' The latter is Anthropic's strongest model built for long-horizon tasks involving complex coding agents. However, the employee also said that this conclusion still awaits more in-depth evaluation and verification.
Early versions of Argon generally received positive reviews in internal testing. But some employees also pointed out that the early versions still struggled with some programming tasks, and their level was roughly equivalent only to Anthropic's earlier-released Claude Opus 5. This feedback indicates that there is still room for improvement in the actual capabilities of Google's new model in coding scenarios.
It is worth noting that Google's internal product codenames and final public release names are often not linked. Internal documents record that the model previously selected to be released externally under the name 'Argon' actually had the internal codename 'Barium-B.' Therefore, whether Carbon will be launched as a subsequent fine-tuned version of Argon or appear as an independent new model under the Gemini 4 family is still undecided.
From the competitive landscape perspective, the advancement of the Gemini 4 series is seen by outsiders as a key step for Google to catch up with Anthropic and OpenAI. If Carbon's coding capabilities are ultimately verified, Google is expected to gain more share in the developer-facing agent market. Conversely, the shortcomings of early versions in programming tasks could also affect the pace of deployment in enterprise applications.
Currently, Google has not yet announced the public release time or specific capability parameters for Argon. As internal testing continues to advance, Carbon's final positioning and naming may still change, and the industry is closely watching Google's next move in the next-generation large model race.
