Quiet admission of the week, dressed up as a policy tweak. Google has started giving all of its engineers access to Anthropic’s Claude Opus 5 for coding, through its internal development platform Antigravity. This is the company that builds Gemini, one of the world’s leading models, deciding that for actual day-to-day engineering its people should also be able to reach for a competitor’s model. The subtext writes itself.
What it says, and what Google insists
Google was quick with the face-saving line: “Gemini remains our primary and foundational model for internal development.” Access to Claude comes with per-employee quotas and is framed as a complement, not a replacement. Previously Google had blocked most staff from external coding tools like Claude Code and OpenAI’s Codex, with exceptions only for some DeepMind teams and high-priority projects. Opening the door to everyone is a real reversal.
Read past the PR and the signal is striking. If Gemini were unambiguously the best tool for every coding job, Google would not need to let its engineers use Claude at all. That it does tells you Claude Opus 5 is good enough at code that even its direct rival would rather its engineers do well than stay loyal to the in-house option. In a competitive market, letting your own team use the competitor is the most honest benchmark there is.
The takeaway
For anyone choosing an AI coding tool, this is a useful tell: Anthropic’s Claude has a genuine edge on code, endorsed by the unlikeliest possible customer. For the industry, it is a reminder that the model wars are being decided by usefulness, not loyalty, and that even the giants will pick the better tool when real work is on the line. Pride is expensive when your engineers’ output is the thing being measured, and Google just decided it was not worth paying.
Related: Claude vs ChatGPT vs Gemini.