Google has spent billions pushing its engineers to use Gemini. So the fact that it’s now letting them use Claude instead is worth paying attention to. According to Seoul Economic Daily, Google has quietly opened access to Anthropic’s Opus 5 model through Antigravity, its internal development platform, giving engineers a path to Claude that simply didn’t exist before.
Until recently, tools like Anthropic’s Claude Code and OpenAI’s Codex were off-limits for most Google staff. Engineers were expected to build with Gemini. Only select teams at Google DeepMind and a handful of high-priority engineering projects had clearance to use outside models. That policy has now shifted, at least partially, with Claude available under per-user quotas through the Antigravity platform.
Google’s official line is measured. The company says Gemini remains its primary internal model and that third-party access is quota-limited and designed to complement Gemini rather than replace it. But that framing doesn’t fully explain why the policy changed in the first place. The more direct explanation: engineering velocity became a bigger priority than model loyalty. When developer productivity is the metric, and Claude outperforms your in-house tool on coding tasks, the calculus shifts.
That’s the uncomfortable read here. Google recently updated its Flash 3.8 model, but on coding benchmarks Anthropic and OpenAI still lead. Some Google employees have reportedly expressed frustration with Gemini’s limitations in practice. The company also released Gemini CLI, an open-source coding agent, but analysts generally view it as weaker than Claude Code on reliability and consistent output, even if it offers a larger context window and generous free tier.
Google isn’t alone in making this call. Amazon, which has its own Nova model family, reportedly allowed employees to use both Claude and OpenAI’s Codex earlier this year. Both Amazon and Google are major Anthropic investors. Google has committed up to $40 billion in Anthropic; Amazon’s total commitment reaches up to $33 billion. So while the optics are awkward, the financial relationship between these companies is already deeply entangled.
What this signals for the broader AI coding market is real. A few things stand out:
- Claude Code has become a credible internal benchmark, not just a consumer product
- Model loyalty programs inside Big Tech have limits when productivity gaps are visible
- Anthropic is winning developer trust even inside organizations building direct competitors
- Enterprise AI tool adoption may increasingly be driven by output quality, not vendor alignment
For developers and founders evaluating AI coding tools, this is useful signal. When the company that makes Gemini starts routing engineers through Claude, the performance gap isn’t theoretical. It’s operational. That’s the kind of validation no press release can manufacture.
