QWERTY Bench ranks AI models by the distance it takes to type their names. One imaginary finger travels between keys on a fixed keyboard layout. We add up that journey, give it a bar, and pretend that a longer trip proves something impressive about the model.
A short name can beat a longer one if its letters sit far apart. Repeating the same letter adds no distance at all. Spaces can send the finger down to the space bar and back, giving elaborate product names another opportunity to look accomplished.
How QWERTY Bench measures typing distance
Each supported character has a key-center position on an approximate US QWERTY grid. The number row sits above three staggered letter rows, and the space bar has its own position below them. A distance of one key width means moving between two neighboring keys on the same row.
For each consecutive pair of keys, we measure the straight-line distance between their centers. Horizontal and vertical movement both count. We add those distances and round the final total to one decimal place. There is no travel before the first key or after the last one, so a single-character name scores zero.
Which characters count on the keyboard
We use the displayed product name, with evaluation settings removed in the same way as Character Bench. Letters, digits, spaces, and supported punctuation all contribute to the path. Uppercase letters use their lowercase key positions; pressing Shift adds no extra journey.
Shifted symbols share the position of their underlying key, so an exclamation mark uses the 1 key. Typographic dashes use the hyphen key. A character absent from the layout breaks the path, and the next supported character starts a new segment. We do not invent keystrokes for emoji or accented letters.
Comparing names and sharing the result
Start with leading generations from OpenAI, Google, Anthropic, and xAI, or choose up to 16 entries from the full collection. Bars sort by travel distance. Download the chart to share the result with the franklineh.com credit included.
The nightly catalogue supplies the names, so newly collected models and naming changes can alter this leaderboard. The score measures an imaginary typing route. It cannot predict typing speed, model quality, or the answer to your next prompt. A name that makes the finger cross the keyboard has simply won this particular joke.
Explore the other ridiculous benchmarks or compare model performance in our AI benchmarks.