Report
Gemini 4 Argon (high) reportedly scores 20.2% on WeirdML v3
The benchmark's creator says that puts it between Sonnet 5.5 and Fable 5.1, but behind the frontier on token efficiency.
TLDR
The creator of WeirdML v3 says Gemini 4 Argon (high) scored 20.2% on the benchmark, between Sonnet 5.5 and Fable 5.1. They call it a strong result but say it remains behind the frontier, especially on token efficiency. WeirdML v3 has 11 hand-made tasks that ask models to explore unfamiliar data, build machine-learning and data-analysis pipelines, and produce results despite limited data, unspecified goals or very limited feedback.
Combined views
—
2 Sources, first seen ago
