Google built Gemini to be the center of its AI strategy. Now it is giving more of its own engineers access to a rival model.

The company has expanded access to Anthropic’s Claude Opus 5 for internal coding through Google’s Antigravity development platform, according to Business Insider, which cited employees familiar with the change. A Google spokesperson confirmed that engineers can use select third-party models alongside Gemini for specialized work.

The shift does not mean Google is abandoning Gemini. Claude access is limited by quotas and routed through Antigravity, while Google continues to describe Gemini as its primary model for internal development.

The decision marks a change from Google’s previous approach, which generally restricted employees from using external coding tools such as Anthropic’s Claude Code and OpenAI’s Codex. Access had been available to some Google DeepMind teams and certain high-priority engineering projects.

Google is not, however, giving employees unrestricted access to Claude Code itself. Claude is available inside Antigravity and subject to individual usage quotas.

“Engineers have access to select third-party models in Antigravity, which is aligned with our external Antigravity enterprise offering,” a Google spokesperson said, according to Business Insider. “Gemini remains our primary and foundational model for internal development, with third-party models available on a quota to support specialized use cases.”

Coding pressure builds

The change comes as AI coding becomes one of the most competitive areas in the model industry. Anthropic and OpenAI have gained attention for their coding systems, while some Google engineers have reportedly complained about Gemini’s performance on difficult programming tasks.

Google has continued updating its Gemini Flash models, but the company is also working to strengthen its coding capabilities as competition increases. That puts Google in an unusual position. It is asking engineers to build with its own AI while simultaneously giving them access to a competitor when Gemini is not the right fit.

Rather than forcing employees to choose Gemini regardless of the task, Google can observe how another leading model performs against its own systems in real development environments. Engineers can also compare models based on the work they actually need to complete, rather than relying solely on public benchmarks.

That could give Google a practical source of feedback for improving Gemini. At the same time, the quota system limits how far employees can move away from Google’s own models. Gemini remains the default, while Claude is positioned as a supplemental tool for specific jobs.

Google competes with Anthropic — and invests in it

The arrangement is particularly notable because Google is both a competitor to and investor in Anthropic. Google announced plans earlier this year to invest as much as $40 billion in the Claude maker, TechRepublic’s Liz Ticong previously reported.

That makes the internal Claude rollout less unusual than it might first appear. Large technology companies are increasingly competing at the model level while still using or supporting rival systems where they offer practical advantages.

Amazon has taken a similar approach by allowing employees to use outside AI coding tools while continuing to develop its own models and AI services.

More Google coverage

What it means for developers

The immediate benefit is greater flexibility for Google’s engineers, but quotas and the requirement to use Claude inside Antigravity limit how freely it can replace existing workflows.

For Google, the move reflects a practical calculation: when AI coding can affect engineering speed across thousands of employees, restricting workers to a single model may matter less than giving them access to whichever approved system performs a particular job well.

It also points toward a workplace where companies may use several AI models rather than committing every task to one provider. Google is still putting Gemini first, but allowing Claude through its own development environment shows that model loyalty has limits when engineering productivity is at stake.

That could become a common enterprise pattern. Instead of standardizing every workload on one AI model, companies may standardize the security, access, and governance layer while allowing developers to choose from several approved models underneath it.

Google is still putting Gemini first. But its willingness to make Claude available internally suggests that, for software development, the winning strategy may be less about choosing one model and more about controlling how multiple models are used.

Let us teach you How to Talk to AI for free! Try our six-minute course at The Neuron Academy and learn a few simple ways to write better prompts and get more useful results from AI, or browse our other AI course for free for seven days. Check out all the lessons here →