Claude got into Google, but only through Antigravity

Google has announced plans to put up to $40 billion into Anthropic, and until recently most of its own engineers still could not touch Anthropic's models at work. That has changed: Business Insider reports, citing two employees and a company spokesperson, that engineers across Google can now use Claude Opus 5 for internal work.

At a glance

• Google has generally kept employees off outside coding tools such as Claude Code and OpenAI's Codex; DeepMind teams and a few high-priority projects were the exceptions, according to three people.
• Access runs only through Antigravity with per-user quotas, and the spokesperson said Gemini remains the primary and foundational model for internal development, with third-party models on quota for specialized cases.
• Despite recent gains from its Flash 3.8 model, Google still trails Anthropic and OpenAI on coding, and the spokesperson described Claude as supplementary to Gemini rather than a replacement for it.

If you missed the earlier episodes: by April 2026 the Los Angeles Times reported that worry about the AI coding race was mounting inside DeepMind, that Gemini's coding features were spread across half a dozen products with different branding, and that some Google engineers already preferred Claude Code. The same reporting said DeepMind had formed a team devoted to AI coding, led by research engineer Sebastian Borgeaud.

Opus 5 is allowed, but only inside Antigravity

Staff across Google may use Anthropic's Opus 5 for internal work, and only within Antigravity, Google's internal version of its development platform. The model arrived with per-user quotas, and the spokesperson described it as supplementary to Gemini.

The spokesperson also said engineers have access to select third-party models in Antigravity, an arrangement aligned with Google's external Antigravity enterprise offering. Antigravity itself is not new: according to the Antigravity site, Google introduced it on 18 November 2025 and has updated it through 2026 with new Gemini model versions, custom agents and enterprise integration. On Gemini's standing, the spokesperson's wording left little room:

Gemini remains our primary and foundational model for internal development, with third-party models available on a quota to support specialized use cases.

DeepMind had Claude while the rest of Google did not

Google has typically kept most employees away from outside coding tools such as Claude Code and Codex, expecting them to build with Gemini instead. Teams inside Google DeepMind have had Claude for some time, and some employees secured it for high-priority engineering projects, according to three people with knowledge of the matter.

Alphabet said in February that roughly half of new code at the company is written by AI, the Los Angeles Times reported. The split in tool access irritated engineers, because Google was also pressing staff to show they were using AI to be more productive, as Business Insider previously reported. Amazon went through a version of this earlier this year, opening Claude and Codex to employees after staff complaints.

Antigravity ships as a platform with a CLI, IDE extensions and an SDK

Google describes Antigravity as a platform built for developers in the agent-first era, shipping as Antigravity 2.0, an Antigravity CLI, IDE extensions and an SDK. For enterprise customers it comes with eligible Gemini Enterprise app subscriptions, with administrative and spend controls out of the box and extensions that put it inside IDEs like VS Code, according to Google Cloud's blog.

The governance layer is where the quotas live: granular spend thresholds, pooled quotas that let a shared token pool absorb capacity instead of leaving purchased quota idle, overage enablement, and centralized metrics tracking token consumption and API calls. A per-user quota, in plain terms, means each engineer's usage is capped and counted rather than drawn from an open pool.

The Los Angeles Times reported in April that Chief AI Architect Koray Kavukcuoglu was working with Google's main engineering team to unite the company's internal AI coding tools under Antigravity in the coming weeks.

Anthropic says Opus 5 more than doubles Opus 4.8 on Frontier-Bench v0.1

In its own description, Anthropic puts Opus 5 at state-of-the-art results on coding and knowledge-work evaluations such as Frontier-Bench and GDPval-AA, while it remains behind Mythos 5 on cybersecurity tasks. On Frontier-Bench v0.1, Anthropic says the model surpasses all others and more than doubles Opus 4.8's performance at a lower cost per task.

Anthropic also states Opus 5 delivers greatly improved performance for the same cost as Opus 4.8, and that it is the new default model on Claude Max and the strongest model on Claude Pro. These are Anthropic's own figures on Anthropic's own evaluations.

How large the per-user quota is, which other third-party models count as "select", and whether Claude Code itself stays barred outside Antigravity are all undisclosed. The proportions are odd, in our view: up to $40 billion committed in one direction, and access coming back measured by the seat. Read next to the spokesperson's wording, the setup looks like a controlled top-up rather than a change of default tools.

Whether the quota loosens

Google has not said how large the per-user allowance is, whether it differs by team, or whether the arrangement carries an end date. Nor has it said whether Claude Code and Codex stay barred outside Antigravity. The thing to watch is the next Gemini coding release: the spokesperson's line keeps third-party models on quota for specialized cases, so a widening of that quota, or a quiet narrowing, is the signal that the internal balance has moved.