Showcase · Karpo Decision vs. a general-purpose LLM
36 events, one request.Which model answers first?
Same request, sent at the same moment. Karpo Decision answers all 36 questions in one forward pass; the LLM generates its answers token by token.
Tonight's criteria
Karpo Decision 31B
One forward pass · a probability for every option
Ready · open a poster to read its listing
General-purpose LLM
General-purpose LLM · generates its answers token by token
Ready · open a poster to read its listing
The city's ready when you are.
Both start together · each clock stops when its own answer arrives
How the comparison works
Both models get one and the same request: the criteria and the 36 listings as text, in English whatever the page's language, and one question per event with three options: match, ruled out, needs more information. The LLM gets it with one instruction more, to reply with one option per question. Neither model sees the posters; they are for you.
Karpo Decision returns its choice and a probability for every option. The LLM returns the option it wrote, without a probability for each option.
The large clocks are wall time in your browser, from the moment the scan starts to the moment each answer arrives, network included. The time under each grid is our server's round trip to the model's host. The rules on this page are the answer key: they check each listing against the criteria, and a model's answer that differs from them is marked.
The events are fictional and the posters AI-generated. The date is fixed to 13 November 2026, and distances are from a demo starting point.