Gemini 3 Pro

Real-world software engineering tasks from GitHub issues. Models must diagnose bugs and write patches that pass test suites. Human-verified subset of SWE-bench.

Unusual and adversarial machine learning challenges. Tests robustness of reasoning about edge cases in ML systems.

Complex terminal-based engineering tasks. Models must use command-line tools, navigate filesystems, and debug systems through shell interaction.

Stanford HELM WildBench evaluation. Tests reasoning on challenging real-world tasks.

Abstraction and Reasoning Corpus. Tests fluid intelligence through novel visual pattern recognition puzzles. Core measure of general intelligence.

Deceptively simple questions that humans find easy but AI models often get wrong. Tests common sense and reasoning gaps.

Mock AIME (American Invitational Mathematics Exam) problems from OTIS. Tests mathematical competition performance.

Original research-level math problems created by professional mathematicians. Problems are unpublished and cannot be memorized.

Stanford HELM evaluation of mathematical reasoning across diverse problem types.