Google has made Anthropic's Claude Opus 5 available to all of its engineers through its internal development platform, Antigravity, according to a Business Insider report — a striking break with precedent at the company that built Gemini, and a candid acknowledgment that the best coding model for its own developers is currently made by a rival.
Previously, Claude access inside Google was restricted to select teams within Google DeepMind and a small number of high-priority engineering projects. Now any Google engineer can select Opus 5 inside Antigravity, the company's agentic development environment. For more context on this story, see our ongoing AI news.
A Google spokesperson downplayed the significance, telling the outlet: "Engineers can access select third-party models in Antigravity, consistent with our external Antigravity enterprise offer."
Claude, With Strings Attached
The access comes with clear guardrails. Google engineers cannot use Anthropic's Claude Code CLI tool directly; they must work with the Opus 5 model inside Google's own Antigravity platform. Each employee is subject to an individual usage quota, and Google has reportedly positioned Claude as a complement to Gemini rather than a replacement, with Gemini remaining what the company describes as its "primary and foundational" model for internal development.
The restrictions say as much as the access. Google wants its engineers productive more than it wants them exclusive — but it also wants their usage data, workflows and feedback flowing through Google's own tooling rather than a competitor's.
Why Google Blinked: Frustration Over Gemini's Coding
The policy change did not emerge from nowhere. Multiple sources familiar with the matter told Business Insider that discontent over Gemini's coding performance has been building inside Google for some time.
The company has shipped Flash-series updates at a remarkable pace in recent weeks — Gemini 3.6 Flash on July 21, 3.7 Flash on August 13, and 3.8 Flash on September 2, three generations in roughly six weeks. Despite that cadence, Gemini continues to trail rival flagship models on coding work, which is precisely the use case developers care most about and the one where Claude has built its reputation.
Engineers reportedly complained about the contradiction of being pushed to use AI tools for productivity while being denied access to the most capable external coding tools. Claude access for a handful of teams was seen by many as an arbitrary restriction that hurt both morale and output.
The Gemini 3.5 Pro Problem
The backdrop to all of this is Google's missing flagship. At Google I/O in May 2026, the company previewed Gemini 3.5 Pro with a planned June launch. Months later, it has not materialized. The Pro line has now gone more than half a year without a successor, dating back to Gemini 3.1 Pro in February.
According to earlier reporting by The Wall Street Journal, some Gemini 3.5 Pro candidate models failed to demonstrate sufficient performance advantages over the Flash series during development and were scrapped. Google is reportedly concentrating engineering and compute resources on improving coding capabilities and increasing investment in reinforcement learning.
Meanwhile, co-founder Sergey Brin has been personally pushing the Gemini team to accelerate iteration, with a particular focus on recursive self-improvement, according to an August Reuters report. Screenshots circulating in the AI community recently showed a model label reading "rsi-model-liverl-le" appearing in Google API responses, fueling speculation that such experiments are already running — though Google has not confirmed anything.
An Awkward Balance With a $40 Billion Partner-Rival
Google and Anthropic share one of the most complicated relationships in the industry. Google announced plans earlier this year to invest up to $40 billion in Anthropic, making it one of the startup's largest backers even as the two companies compete directly for the same customers and, now, the same internal users.
Anthropic, for its part, is reportedly preparing for a public offering at a valuation of roughly $2 trillion. Having its flagship model adopted broadly inside Google — the company that is simultaneously its investor and its competitor — is a powerful endorsement of that valuation logic: enterprises and giants alike are choosing Claude where coding capability matters most.
Google is also not the first tech giant to make this pivot. Amazon began allowing employees to use Claude and OpenAI's Codex coding tools earlier this year, according to prior reports, suggesting a pattern: even companies investing billions in their own models are conceding that no single lab leads at everything.
What It Signals for the AI Coding Race
AI coding tools have become the central battleground for model companies competing for developer mindshare. Anthropic's Claude Code and OpenAI's Codex command large followings among developers, while Gemini's reputation in coding has lagged its strengths in other modalities.
Google's decision is pragmatic in the short term and revealing in the long term. Pragmatic, because unlocking engineers with the best available tool is obviously better than hobbling them with a quota-protected rival. Revealing, because when the company with the largest AI research organization on earth decides its own engineers should reach for a competitor's model to write code, the capability gap it describes internally is not a marketing problem — it is a product reality.
The pressure now falls squarely on the delayed Gemini 3.5 Pro. Until it ships and closes the gap, Google will keep paying — in quotas, in awkward headlines, and in market share — for the coding advantage it has not yet built.
For ongoing coverage of the models, the money and the tools shaping the AI industry, AI Buzz Wire follows every move.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
