AI Benchmarks

CursorBench: Real-World IDE Code Editing and Navigation

Practical benchmark measuring an AI model's ability to perform code edits, fast diff generation, symbol resolution, and context retrieval in realistic editor workflows.

SWE-bench: Resolving Real-World GitHub Issues (Verified Subset)

Standard benchmark evaluating an AI model's ability to solve end-to-end software engineering problems from real GitHub issues in popular Python repositories.

LiveCodeBench: Holistic and Contamination-Free Code Evaluation

Continuously updated programming benchmark collecting problems published after model training cutoffs from LeetCode, AtCoder, and Codeforces to prevent data contamination.

LiveCodeBench Pro: Frontier Hard Algorithmic Benchmark

Hard subset of LiveCodeBench focusing on advanced competitive programming problems (Div 1 / Div 2 Codeforces and Hard LeetCode).

Humanity's Last Exam: A Benchmark for Frontier AI

A rigorous multimodal benchmark developed by CAIS and Scale AI designed to be resistant to saturation, consisting of 3,000 expert-level questions spanning dozens of academic and professional fields.

MMLU-Pro: A More Robust and Challenging Multi-Task Benchmark

An enhanced version of Massive Multitask Language Understanding (MMLU) featuring 10-choice questions and a higher proportion of reasoning-focused problems.

Massive Multi-discipline Multimodal Understanding

Multimodal benchmark assessing visual reasoning across 30 college-level subjects with diagrams, charts, medical scans, maps, and technical illustrations.

Google-Proof Q&A (Diamond Subset)

Graduate-level physics, chemistry, and biology questions written by domain experts. Non-expert humans with unrestricted web access score only 34%.

American Invitational Mathematics Examination

Prestigious high school mathematics competition problems requiring complex multi-step creative deduction and integer answers from 000 to 999.

Abstraction and Reasoning Corpus for AGI

Benchmark created by François Chollet measuring an AI system's ability to acquire new skills and solve novel visual grid puzzles with minimal priors.

Berkeley Function Calling Leaderboard

Comprehensive evaluation of LLM tool-calling and function-calling capabilities across simple, multi-turn, parallel, and executable API scenarios.