Notable results
Standouts across all 898 answers. Click any card to open that model in the browser.
Who walks? By manufacturer
Each pie shows one manufacturer's answers, grouped across every host and variant it's served under. Sorted by share of walk answers. Click a pie to browse that manufacturer's answers.
Frequency by manufacturer
Every manufacturer with at least 2 answered models. Bar length is always 100% of that manufacturer's answers. The right column shows walk % and model count.
Models per manufacturer
How much of the catalogue each manufacturer accounts for, answered models only.
By host provider
Puter's upstream routes. Often the same model gives different answers depending on the host.
Walking over time
Answers grouped by model release date, in half-years. Column height is the number of models; the label above is the walk share.
What predicts a walk?
Walk vs. drive split across model traits. Each row is 100% of the models with that trait.
Output price ($ / 1M tokens)
Reasoning, weights, variant
How they answered
Length, speed and cost of each response, split by verdict. Values are medians, with p90 in parentheses.
Same model, different answer
Base models served by more than one host or variant that didn't agree with themselves. Sorted by how evenly split they are.
Newest models that still walk
The most recently released models that said to walk.
In their own words: walk
A rotating sample of walk answers, quoted by the judge.
Models that didn't answer
Calls that failed or were skipped. None were caused by Puter's own rate limits.