Gemini 4 Argon's 97.5% on ARC-AGI-3

The headline score comes from a custom setup. In the standard harness it falls to 49.8%, so most of the gain comes from Google's scaffolding, which builders do not get by calling the API. Argon also trails rivals in math and software engineering, the tasks most agent products depend on. Anyone choosing a model should test it on their own tasks rather than plan around this number.

Read the coverage (The Decoder)

Companies in the story: GOOGL

A view on what the development means for people building products. Not a view on any company's stock and not investment advice.

Every Not this Today's brief