Companies·3 min read
By BitsMindsSource: Business Insider

Google Lets Every Engineer Use Claude Opus 5

Anthropic’s Opus 5 is now available to engineers across Google, but only inside Antigravity, the company’s internal build of its development platform. Access had been limited to parts of DeepMind and a handful of high-priority projects. Gemini stays the primary model, and Claude arrives on a per-user quota.

Google opens Claude Opus 5 to every engineer A Google employee access badge hangs on a lanyard between a large Google mark and a large Anthropic Claude starburst. The badge reads Opus 5 and shows a partly filled quota meter, an editorial metaphor for per-user access inside Google's Antigravity tool rather than a depiction of a real badge. Opus 5 PER-USER QUOTA ANTIGRAVITY Gemini still comes first. Claude is now allowed. BITSMINDS.COM
Share:

Google has made Anthropic's Claude Opus 5 available to engineers across the company, Business Insider reported on Tuesday, citing two employees and a confirmation from a company spokesperson. The access is narrower than the headline suggests: staff can select Opus 5 inside Antigravity, Google's internal build of its own development platform, and each person gets an individual usage quota. Anthropic's Claude Code remains off the menu.

The policy it replaces was unusually strict for a company of Google's size. Most employees have been barred from outside coding tools — Claude Code and OpenAI's Codex among them — and expected to build with Gemini. There were exceptions: teams inside Google DeepMind have had Claude for some time, and individual engineers secured it for high-priority projects, according to Business Insider's sources. What changed is that the exception became the default.

Google's framing is careful. "Gemini remains our primary and foundational model for internal development," a spokesperson told the publication, adding that third-party models are offered on a quota to support specialised use cases and that the arrangement mirrors the external Antigravity enterprise product. In other words: not a migration, a supplement — metered, and inside Google's own tooling, where Google keeps the telemetry and the billing relationship.

The reason is the part nobody at Google has to say out loud. Coding is the one frontier capability where Gemini has not caught up, and internal frustration has been building even as leadership pressed staff to demonstrate that AI is making them faster. BitsMinds scored Opus 5 against GPT-5.6 Sol in a one-prompt build-off on 11 September, and put the three big agentic coding tools — including Antigravity — head to head in our tooling comparison. The gap Google is buying around here is not a rumour inside the company; it is a daily experience for the people shipping its code.

There is a financial circularity to it as well. Google is one of Anthropic's largest backers and said earlier this year that it would invest up to $40 billion in the company. Paying for a rival lab's model, through a product Google owns, funded partly by Google's own money, is a strange loop — but it is the same loop every hyperscaler is now running, and it is cheaper than a year of lost engineering throughput.

Google is also not the outlier. Amazon began letting employees use Claude Code and Codex earlier this year after similar complaints. Meta went in the opposite direction in June, barring its engineers from both tools over distillation concerns. Three companies, three readings of the same trade-off between protecting a model programme and unblocking the people who use one.

The detail worth holding onto is the quota. A blanket ban is a statement about strategy; a per-seat allowance is a budget line, reviewed quarterly, that rises or falls with how far behind Gemini is on any given month. Google has not conceded the coding race. It has started paying rent in it.

More on Claude

Evergreen coverage we keep current — start here.

Want AI news before everyone else?

The morning's most important AI stories, straight to your inbox. No fluff.

Related Articles

Microsoft AI code of conduct: never resist shutdown A human hand presses a physical power button beside a calm, inactive robot with the Microsoft mark. A cream paper document titled AI Code of Conduct is visibly stamped Draft and lists accepting correction, allowing shutdown, and communicating clearly. The image is an editorial metaphor for the proposed policy, not a depiction of Microsoft hardware or a claim that compliance has been verified. Microsoft AI / HUMAN CONTROL DRAFT CODE OF CONDUCT STANDBY MAI HUMAN CONTROL Microsoft DRAFT AI Code of Conduct 01 Accept correction 02 Never resist shutdown 03 Communicate clearly PUBLIC FEEDBACK Six weeks before finalisation The off switch stays human. BITSMINDS.COM
Companies

Microsoft's AI Code of Conduct: Never Resist Shutdown

Positron Raises $875M for Memory-First Inference Chips
Companies

Positron Raises $875M for Memory-First Inference Chips

SAFETY & SECURITY COMMITTEE Paul Christiano BITSMINDS.COM
Companies

OpenAI Puts Paul Christiano on Its Foundation Board