Googlers finally get Claude, but only inside Antigravity

For years the rule inside Google was to build with Gemini and leave Claude Code and Codex alone. According to Business Insider, engineers across the company can now use Anthropic's Opus 5 for internal work, but only inside Antigravity.

At a glance

• Two employees described the change to Business Insider and a spokesperson confirmed it: one of Anthropic's strongest coding models is now available to engineers across Google through its internal development tool.
• The model is Opus 5, the venue is Antigravity, and the limit is a per-user quota; Google is also a backer of Anthropic, with plans announced this year to invest up to $40 billion.
• Gemini stays the default: the spokesperson calls it Google's primary and foundational model for internal development, with third-party models on a quota to support specialized use cases.

If you missed the earlier episodes: Business Insider reported that in June 2025 engineering vice president Megan Kacholia emailed engineers telling them to use AI tools in their coding, with updated guidelines saying engineers should use only internal models for coding. At a July all-hands, Sundar Pichai told employees they need to use AI for Google to lead the race. By April this year, the Los Angeles Times reported leaders were anxious about falling behind, and some engineers preferred Claude Code.

Opus 5 arrives with per-user quotas and one place to run it

Two employees told Business Insider about the change, and a Google spokesperson confirmed it. Staff across the company may now use Opus 5 for internal work, inside Antigravity and nowhere else. Google has typically barred most employees from outside coding tools such as Claude Code and OpenAI's Codex, expecting them to build with Gemini instead.

The spokesperson said engineers have access to "select third-party models in Antigravity, which is aligned with our external Antigravity enterprise offering", and that Gemini "remains our primary and foundational model for internal development, with third-party models available on a quota to support specialized use cases". Claude comes with per-user quotas and is meant as a supplement to Gemini.

Anthropic's own pitch for Opus 5: state of the art on coding and knowledge-work evaluations such as Frontier-Bench and GDPval-AA, with results on Frontier-Bench v0.1 that, the company says, beat every other model and more than double Opus 4.8's performance at a lower cost per task.

Google is putting up to $40 billion into Anthropic

Google is a backer of Anthropic and announced plans earlier this year to invest up to $40 billion in it. It has also spent billions building Gemini and pushed employees toward its own AI tools. On coding, Google is still behind Anthropic and OpenAI, despite recent gains from its Flash 3.8 model.

The gap shows up in how engineers outside Google work. According to the Los Angeles Times, TinyFish co-founder Keith Zhai said engineers in Silicon Valley toggle between Claude Code and Codex to see which gives the better result, and Google often is not even in the conversation. Amazon, after staff complaints, started letting employees use the coding tool and Codex earlier this year.

Gemini's limitations had been grating on some Googlers

Business Insider previously reported that the frustration was sharpened by pressure on staff to show they were using AI to be more productive, while Claude stayed with select teams. Access before this change ran to teams inside Google DeepMind and, according to three people with knowledge of the matter, to individual engineers who secured it for high-priority engineering projects.

According to the Los Angeles Times, Gemini's capabilities were described as sprinkled across half a dozen coding products with different branding, and Chief AI Architect Koray Kavukcuoglu was working with the main engineering team to unite the internal coding tools under Antigravity. For scale: Alphabet said in February that roughly 50% of new code at the company is written by AI, the paper reported.

The internal Antigravity is aligned with the enterprise offering

Antigravity is Google's development platform, and what engineers use is Google's internal version of it. According to Wikipedia, the platform was announced on 18 November 2025 alongside Gemini 3, is a heavily modified fork of Visual Studio Code, runs its coding agents mainly on Gemini 3.1 Pro and Gemini 3 Flash, and already supported Claude Sonnet 4.6, Claude Opus 4.6 and the open-source GPT-OSS-120B.

Google's Codelabs material describes Antigravity 2.0 as a wider set of pieces: a standalone app, an IDE, a CLI, an SDK and extensions for VS Code, JetBrains and Zed. In practice Opus 5 sits as one selectable model inside a tool Google builds and meters. Think of the quota as a mobile data plan: the model is there, but the meter runs against your name.

Neither the spokesperson's statement nor the reporting puts a number on the per-user quota, so "access to Opus 5" could mean a working day of agent runs or a handful of tasks. In our view the shape of the concession is the interesting part: the rival model arrives wrapped in Google's own tool, where Gemini stays foundational and third-party use reads as a specialized case rather than a daily habit.

Whether Claude Code follows The permission is scoped to Antigravity, and no date has been given for widening it to Claude Code or Codex as standalone tools. The checkpoints worth watching are whether the per-user quotas grow, whether the external Antigravity enterprise offering carries Opus 5 on the same terms, and what Alphabet says next about AI-written code after February's roughly 50% figure.