News sentiment
Which Claude model is worth paying for?
Give it news headlines. Each AI model marks every one as good, bad, neutral or irrelevant news for a company, and the page shows which model gets the most right for the least money.
What it's for
News sentiment is a common kind of alternative data: headlines turned into a daily score per company. The raw feed is noisy, with passing mentions, repeated stories and wording like "cuts outlook" versus "cuts costs", so every headline has to be labelled before it is usable. Models do the labelling, and the largest Claude model costs about a hundred times the smallest. This tool measures whether the extra cost buys better labels.
What you upload
A spreadsheet of headlines, one per row, each with the label a person gave it: positive, neutral, negative or irrelevant. Those human labels are the answer key the models are marked against. The demo uses 50 made-up headlines.
A plain list of headlines also works, but without an answer key the page can only compare the models with each other.
How it picks
Every model labels every headline. The page then recommends the cheapest model that is nearly as accurate as the best one, within 2 percentage points, and still catches at least 9 in 10 negative headlines, because those are the ones an investor can't afford to miss. Costs come from what each call actually used, not from estimates.
Cached shows a saved run of the demo for free. Live runs the models for real and needs a passcode.