Google is testing several Gemini 4 variants, including Carbon, which is said to outperform Argon on programming tasks. The model was deployed on Google's internal coding platform Jetski and is being compared to Anthropic's strongest coding model, Opus 5.5.
Internal feedback suggests Carbon is more capable than earlier Argon versions, which were likened to the older Opus 5. However, Carbon still requires further testing to confirm its full potential.
The Gemini 4 Carbon model is built on Google's internal coding platform Jetski and targets programming tasks. Availability for the model is not yet confirmed, but internal documents suggest it could be an update to the Argon model.
"Have you heard of recursive self improvement," said Vedant Misra, a Google Deepmind employee, referring to using AI to build better AI. Misra's comment highlights the rapid model iteration driven by AI self-improvement.
The announcement follows Google's efforts to enhance its AI models, including the Gemini app's new 'Automatic' mode and AI Studio's 'Ultra' mode. Google is working to maximize the capabilities of its Argon model.
Google did not confirm whether Carbon will ship as an Argon update or a standalone model. The company is also preparing for the Gemini 4 launch, though an official date has not been announced.
Source: thedecoder
