Google has extended Anthropic’s Opus 5 coding model to its entire engineering staff through Antigravity, its internal development platform, according to two employees and a company spokesperson who confirmed the rollout to Business Insider. Access had previously been confined to Google DeepMind teams and a handful of employees who secured exceptions for high-priority projects, according to three people familiar with the matter.
The stakes are structural, not cosmetic. Google has spent years steering staff toward Gemini and away from Claude Code, OpenAI’s Codex and every other outside coding tool. Opening the door to a rival lab’s model, even on a metered basis, concedes that internal pressure alone was not closing the coding gap.
A Google spokesperson described the access as bounded and supplementary. “Engineers have access to select third-party models in Antigravity, which is aligned with our external Antigravity enterprise offering,” the spokesperson said. “Gemini remains our primary and foundational model for internal development, with third-party models available on a quota to support specialized use cases.” The spokesperson added that Claude is offered under per-user quotas and is meant to support, not replace, Gemini.
That framing matters because Google is not a neutral customer here. Earlier this year it announced plans to put as much as $40 billion into Anthropic, which makes the Antigravity rollout as much a strategic hedge as a productivity fix: Google stands to gain from Anthropic’s growth even while competing against it on model quality.
The underlying admission is about capability, not policy. Google’s Flash 3.8 model has posted recent gains, but the company is still trailing Anthropic and OpenAI on coding performance, a gap that reportedly left engineers frustrated with Gemini’s limits inside daily workflows. Business Insider had previously reported that Google was pressuring staff to demonstrate AI-driven productivity gains even while restricting access to the tools best suited to deliver them, a contradiction that made the earlier limited rollout harder to defend internally.
Google is not alone in reversing course under employee pressure. Amazon opened internal access to both Claude and Codex earlier this year after similar staff pushback, suggesting that even the largest AI labs cannot fully insulate their own engineers from tools built by competitors.
The quota structure is the detail worth watching. Google has not disclosed usage caps or how it will measure whether Claude access changes output, which means the current rollout reads more like a pressure valve than a strategic reversal. If per-user limits prove too tight for real engineering work, expect renewed internal friction over why a company invested in Anthropic still treats its model as a rationed resource rather than a default tool.
For teams at other large enterprises watching this shift, the signal is less about Claude specifically and more about the ceiling on mandated single-vendor AI tooling: when an internal model demonstrably lags on a core workflow like coding, engineering orgs will route around it, and leadership eventually follows. Any company standardizing on one AI vendor for developer tooling should build a documented exception path before frustration forces one, rather than after.
Reported by Hugh Langley for Business Insider, published September 14, 2026.
