Google opens Claude Opus 5 to every engineer -- but only inside its own IDE, and only on a quota

Google has dropped an internal policy that kept most of its engineers off Anthropic's models, according to reporting from Business Insider's Hugh Langley: any Google engineer can now select Claude Opus 5 inside Antigravity, Google's own agent-first coding platform, rather than the small set of DeepMind teams and high-priority projects that previously had access. Employees still cannot reach for Anthropic's own Claude Code tool directly -- the model is only available chosen from within Antigravity's own interface, and each employee works against an individual usage quota rather than open access.

Google's own position, as relayed in the reporting, is that Gemini remains its primary and foundational model for internal development, with Claude and other third-party models offered on a quota basis to cover cases where engineers judge Gemini's own coding performance insufficient. Google isn't the first major lab customer to make this concession: Amazon opened internal access to Claude and OpenAI's Codex to its own engineers earlier this year, over the same kind of internal resistance to using a competitor's model that Google had maintained until now.

The routing detail is the part worth sitting with: Google isn't handing engineers an Anthropic account. Claude Opus 5 runs inside Antigravity, Google's own platform, alongside Gemini in the same model picker -- meaning Google controls the interface, the logging, and the usage quota for a model it competes with directly. That arrangement lets Google concede a real capability gap on coding without conceding distribution: whatever an engineer builds with Opus 5 still happens inside Google's own tooling, not Anthropic's.