Character Bench ranks AI models by the length of their product names. Letters count, but so do spaces, version numbers, periods, and hyphens. A longer name produces a taller bar. It is a precise way to reward a model for taking up more room on a slide.
The default comparison includes leading generations from OpenAI, Google, Anthropic, and xAI. Choose models to compare other entries in Franklin's collection, including models from additional companies.
What counts as a character
We count the displayed name character by character. Gemini 3.8 Flash has six characters in Gemini, three in 3.8, and five in Flash. Its two spaces bring the total to 16. The period contributes one character, just like a letter. GPT-6 Astra scores 11, including its hyphen and space.
The score follows the label shown beneath the bar, so you can check the calculation yourself. A change in capitalization leaves the total alone. Adding a suffix, a space, or another digit changes it. Unicode symbols count as code points; a symbol represented by multiple code points can contribute more than one.
Keeping product names separate from evaluation settings
Catalogue entries sometimes append reasoning or evaluation settings such as (High), (Max), or (Default Fallback). We remove those settings before displaying and counting the product name. Otherwise, the same named model could gain characters simply because it was evaluated with a different configuration.
Labels that belong to the product, including preview labels and dated releases, remain in the name. We do not shorten Flash to F, remove version punctuation, or quietly drop spaces. The visible label is the input to the score.
Build and share a character comparison
Use the model picker to build a comparison of up to 16 entries, then reset to the flagship selection whenever you want. The chart sorts longer names first. Downloading the PNG keeps the character totals, axis labels, and franklineh.com credit in the image. A high score says exactly one thing: the displayed name contains more characters.
Explore the other ridiculous benchmarks or compare model performance in our AI benchmarks.