Gemini 4 Argon: Google's Enterprise Model for Long-Horizon Tasks

Google has a new top model, and it is pointed at its biggest customers first. Gemini 4 Argon is being introduced for the enterprise tier and for Google Ultra, and the whole pitch is built around jobs that run long.

On the benchmarks Google shared, it comes out ahead in 14 of 19 tests, including 84.2% on GraphWalks 1M token. That is the kind of number that decides whether a model can hold a long chain of work together.

What Happened

Google introduced Gemini 4 Argon for the enterprise class and Google Ultra. The stated focus is long-horizon reasoning and autonomous agents, meaning tasks that keep going across many steps with little human babysitting.

The benchmark results it leads on:

• It wins 14 of 19 tests.
• 84.2% on GraphWalks 1M token.
• 77.9% on DeepSWE v1.1.

Why It Matters

Long-horizon reasoning and autonomous agents are the two things that separate a chat model from one you can hand a multi-step job to. GraphWalks 1M token is a long-context test, so the 84.2% there speaks to how the model holds up when the input gets big.

If you already sit on enterprise or Google Ultra, that is where this lands. Everyone else is watching from the outside for now, with nothing to sign up for yet.