Antigravity 2.0 · third-party models
Google's agentic dev environment will now run Anthropic's models — opt-in, admin-gated, off by default.
Google’s agentic development environment — an IDE, a CLI and a desktop app where an AI agent reads your code, calls tools and proposes changes. The news is that it will now run Anthropic’s Claude models alongside Google’s own, once an administrator switches them on.
What you build with it
- Agent-assisted coding with a choice of model provider
- Teams that want to change model without changing their whole toolchain
- Comparing how different models handle the same codebase
- Enterprise setups where provider routing has to be an auditable policy
What it gets you
- The model becomes a swappable setting rather than a lock-in
- Off by default — enabling it is an explicit admin decision
- Works across the IDE, the CLI and the desktop app
- Surfaces
- Antigravity 2.0 · CLI · IDEs
- Models
- Claude Opus 5.5 · Sonnet 5.5
- Default
- Disabled — admin opt-in
- Also this week
- Gemini 4 Argon GA (13 Oct)
Straight from the source
Watch instead of reading
The fast version, if you do not have time for the whole write-up.
Why it matters
A platform vendor shipping a first-class path to a direct competitor's models is a real signal. It says the model layer is becoming substitutable and the durable product is the harness around it — the context, the tools, the permissions, the review loop.
The harness is the product now
For two years the assumption was that whoever had the best model won the developer tool. This release points the other way: Google is competing on the environment — how code is read, which tools the agent can call, how changes get reviewed — and treating the model as a component the customer may swap.
If you build agent systems, that is the lesson worth taking. The durable engineering is in orchestration, context assembly, tool definitions and guardrails. The model behind it should be a configuration value, because it will change under you.
Off by default is the interesting part
Third-party models are disabled until an administrator turns them on. That is the correct default for an enterprise tool — sending source code to another vendor's API is a decision that belongs to whoever owns the data policy, not to each developer.
It is also a pattern worth copying in your own tools. If your agent can route to multiple providers, which providers are permitted should be an explicit, auditable setting rather than an implementation detail buried in a config file.
Why this one is marked confirmed
Most entries on this radar are assembled from press coverage. This one comes from Google's own Gemini Enterprise release notes, which is the vendor documenting its own product.
The same notes record Gemini 4 Argon reaching general availability for the Gemini Enterprise web app, mobile app and AI developer tools on 13 October.
Reach for it when
- You are building an agent and want the model to stay a swappable configuration value
- You need per-provider routing to be an auditable policy, not a code change
- Your team already works in Antigravity and wants Claude models in that loop
Watch out for
- Enabling a third-party model routes your source code to that vendor — a data-policy decision first
- Model-specific prompt tuning does not transfer cleanly between providers; expect to re-evaluate