Gemini 4 Argon scores 97.5 percent on ARC-AGI-3 but drops to 49.8 percent under the standard harness

Results swing with the test setup

Google DeepMind's new model Gemini 4 Argon scored 97.5 percent on the ARC-AGI-3 benchmark. That makes Google the second lab, after OpenAI, to cross the 95 percent mark.

Under the standard test harness, however, the same model drops to 49.8 percent. In other words, the result varies widely depending on how the test is set up.

Behind on math and coding

According to the report, Argon also trails its rivals in math and software engineering tests. That suggests the model has not yet taken the top spot in the broader capability picture.

ARC-AGI is a benchmark that measures how well an AI model solves visual reasoning puzzles it has not seen before. A high score alone does not mean overall superiority; the test conditions directly affect the outcome.