Explore · Benchmarks

Benchmarks6

A benchmark groups tasks into a suite. Trending is ranked by trials in the last seven days; popular by all-time trials.

6 results

The reference suite for autonomous software engineering on real repositories.

Verifier-checked competition math from AIME through Olympiad problems.

Tool-agent-user interaction under realistic, policy-constrained scenarios.

General assistant tasks that resist memorization and reward real reasoning.

Realistic, reproducible web agents in self-hosted application sandboxes.

Shell-native tasks that measure real command-line problem solving.