AI news · Google · models
Google staff try a Gemini 4 checkpoint called Carbon, BI says
Business Insider says one employee compares the internal model to Claude Opus 5.5 on coding, as Argon surfaces in Google’s Antigravity tool.
Business Insider reported on Friday 9 October 2026 that Google employees are using a Gemini 4 model codenamed Carbon on the company’s internal coding platform, according to documents and screenshots it saw. One staff member said it felt comparable to Anthropic’s Claude Opus 5.5 for coding, while adding that it needed more use.
Google announced Gemini 4 Argon on 30 September and gave cyber-defence partners first access, as we reported. An internal document lists models named Argon, Barium and Carbon, and notes that one called Barium-B became the public Argon. Earlier Argon versions trailed on some coding tasks, one employee told Business Insider; we reported similar staff doubts on 2 October. Google may not release Carbon.
Argon now appears in the model picker of an insiders build of Google’s Antigravity coding tool, with three context settings: 256,000 tokens by default, or 512,000 and 900,000 at about 1.3 and 1.8 times the quota per turn, TestingCatalog reports. Paid API customers and Google AI Ultra subscribers are expected to get Argon after the cyber partners.
For teams choosing a coding model, Argon’s wider release gives them a Gemini 4 model to set against Opus 5.5.
Sources
- Business Insider: Google is about to roll out a new AI model. Employees say they’re testing another that’s way better.. Checked
- TestingCatalog: Gemini 4 Argon hints emerge as Google tests Carbon checkpoint. Checked