| {"id":"swe_bench_verified","name":"SWE-bench Verified","description":"500 human-validated GitHub issues across 12 Python repos. The harness must produce a patch that resolves the issue and passes the project's test suite.","maxScore":100,"unit":"% resolved","sourceUrl":"https://www.swebench.com/"} | |
| {"id":"terminal_bench","name":"Terminal-Bench","description":"Stanford and Anthropic benchmark of agentic terminal tasks. Each task gives the agent a goal and a sandboxed shell; success is measured by a deterministic post-condition.","maxScore":100,"unit":"% solved","sourceUrl":"https://www.tbench.ai/"} | |
| {"id":"aider_polyglot","name":"Aider Polyglot","description":"225 of the hardest Exercism coding exercises across C++, Go, Java, JavaScript, Python, and Rust. Measures whole-file edit-by-diff quality, not just code generation.","maxScore":100,"unit":"% pass2","sourceUrl":"https://aider.chat/docs/leaderboards/"} | |
| {"id":"swe_lancer","name":"SWE-Lancer","description":"OpenAI benchmark of paid Upwork engineering tasks ($1M+ in real bounties). Includes both diff-style fixes and longer feature work judged against the original buyer's acceptance criteria.","maxScore":100,"unit":"% earned","sourceUrl":"https://github.com/openai/SWELancer-Benchmark"} | |