Plain rules

Terms

Last updated August 20, 2026.

The short version: quote our numbers freely and credit them honestly. They are one lab's measurements, not a promise from any vendor.

What these numbers are
DurkBench publishes numbers we measured ourselves. They come from one lab machine, one connection, the app versions listed on Status, and the exact flags listed on How we test.
They are not a promise from Anthropic, OpenAI, or xAI. Do not call any board here the official speed of a model, and do not present it as a vendor statement.
What you may do
Read, quote, screenshot, and link these numbers — at work, in a blog post, in a purchase memo, anywhere. That is why we publish them.
The JSON feeds, the live badge, and the RSS feed are open to everyone — no key, no account. Our numbers only change every hour, so polling faster gains you nothing.
What you may not do
Do not edit a number and keep our name on it.
Do not resell our data as your own benchmark.
Do not use our name or mark in a way that suggests we tested, endorsed, or partnered with your product. Brand has the rest of those rules.
Other companies’ names
Claude, Codex, and Grok are products of Anthropic, OpenAI, and xAI. Those names, and every model name on our boards, belong to their owners.
We use the names only to say which tools we tested. None of those companies sponsors this site, reviews it, or has any say in what it publishes.
No promises
The site and the data are provided as-is, with no uptime promise. The lab machine can be down, a vendor can be down, and a board can be empty.
Runs fail, apps change, and we can be wrong. Read How we test and check the log before you ship a claim.

How to credit us

“DurkBench, test kit v6, 24-hour window.”

Say it came from DurkBench, swap in the window you used, and link the page you took it from.

DurkBench is run by the DurkBench team. If these terms change, the date at the top of this page changes with them.

What these numbers are01