Explore Singularity
© 2026 Singularity Logs
THE BENCHMARK INDEX
Understand what each evaluation actually measures — and what it does not.
Competition mathematics questions from the 2025 AIME.
Competition mathematics questions from the 2026 AIME.
Novel visual puzzles that test adaptation to unfamiliar patterns.
Human preferences from pairwise comparisons of model responses.
Function calling and tool-use correctness across task types.
An evaluation of evasive language detection in earnings-call answers.
Multi-step information tasks involving reasoning and tool use.
Graduate-level science questions designed to challenge expert reasoning.
Mathematical word problems that require several reasoning steps.
Scenario-based evaluations of information use in long inputs.
Challenging competition mathematics from the Harvard–MIT tournament.
Expert-written questions spanning many academic disciplines.
Continuously refreshed programming problems for code-generation evaluation.
Mathematical problem solving across several competition topics.
A more demanding knowledge and reasoning test across academic subjects.
Questions combining images with knowledge from multiple disciplines.
Real repository issues evaluated with executable software tests.
Practical tasks completed by agents in a terminal environment.
