# DurkBench DurkBench is an independent benchmarking lab. We time Claude Code, Codex, and Grok Build — the paid command-line apps people type into. Not the hidden HTTP APIs. We run every test ourselves, on our own dedicated lab machine, every hour, on subscriptions we pay for. This site publishes the results. Readers do not install, run, or host anything. Test kit v6. Boards use only current-kit rows that passed the answer check. One lab machine. These exact flags. Every published number traces back to a saved CLI file in our archive. These are our own measurements. They are not official numbers from Anthropic, OpenAI, or xAI. ## For people - https://durkbench.com/ — live pulse board - https://durkbench.com/start — how to read the boards - https://durkbench.com/today — the last 24 hours in one card - https://durkbench.com/compare — two models side by side - https://durkbench.com/models — one page per paid model - https://durkbench.com/design — one landing-page brief, seven models, every page they wrote - https://durkbench.com/method — exact prompts, flags, and units - https://durkbench.com/logs — every sample and what our lab spent - https://durkbench.com/status — is our lab healthy - https://durkbench.com/changelog — every change to how we test - https://durkbench.com/about — who runs the lab - https://durkbench.com/teams — using the data before you buy seats - https://durkbench.com/developers — the JSON feeds - https://durkbench.com/privacy - https://durkbench.com/terms ## For machines - https://durkbench.com/api/today - https://durkbench.com/api/speed?hours=24 - https://durkbench.com/api/probes?hours=24 - https://durkbench.com/api/runs?limit=100 - https://durkbench.com/api/status - https://durkbench.com/badge - https://durkbench.com/feed.xml ## Units - Wall tok/s is the fair speed compare: all generated tokens, including think, over start-to-finish time. - Ping is a tiny "ok" check. It is not a speed test. - A clean round is one where the app answered and the answer was right. Only clean rounds set a speed number. - User reports (when enabled) are visitor votes about how a tool feels. They are opinion, kept apart from lab numbers. ## If you quote us Say the number came from DurkBench, name the test kit version, and link the page you took it from.