Google Gemini 4 Leak Reveals Carbon and Flash Models.
Google may be preparing several new AI models under its Gemini 4 family. Recent reports suggest that the company is testing an advanced coding model internally, alongside other unreleased versions.
The model, reportedly codenamed Carbon, has attracted attention because Google employees have praised its coding capabilities. Some internal feedback even compares its performance with Anthropic’s Claude Opus 5.5.
However, these claims have not yet been independently verified. Google has also not confirmed whether Carbon will become a publicly available product.
Google Tests Gemini 4 Carbon for Coding.
According to Business Insider, internal Google documents, employee chat records and screenshots point to ongoing tests of new Gemini 4 models.
Carbon is reportedly available to employees through Google’s internal programming platform, Jetski. Early feedback suggests that the model performs well in software development tasks.
One employee reportedly described Carbon’s coding performance as being close to Claude Opus 5.5. Another employee also praised the model in an internal discussion.
These comments have raised expectations for Google’s next generation of AI models. However, employee feedback alone does not establish that Carbon matches Claude across independent coding benchmarks.
Gemini 4 Argon, Barium and Carbon Explained.
Reports indicate that Google has used several internal codenames while developing its Gemini 4 models.
The three names attracting the most attention are Argon, Barium and Carbon.
Argon: The reported name of Google’s Gemini 4 flagship model.
Barium: An earlier internal development series. Reports suggest that Barium-B was selected for release under the Argon name.
Carbon: A newer model reportedly being tested for programming and coding tasks.
These codenames do not necessarily correspond directly to final product names. Google can test several versions before deciding which model to release.
Therefore, it remains unclear whether Carbon will become an upgraded version of Argon or launch as a separate model.
Can Gemini 4 Carbon Compete With Claude Opus 5.5?
Carbon’s reported coding performance is particularly interesting because early feedback about Argon highlighted some differences between Google’s models and Anthropic’s Claude.
According to the report, Argon received positive feedback in several areas. However, some testers reportedly identified weaknesses in certain programming and terminal-based tasks.
