Model tracker: decision models compared on price, speed and benchmarks

What's a decision model? A model that answers a typed question (yes or no, pick one, pick several, a number) with probabilities instead of writing text, in milliseconds and for a fraction of an LLM call. Agents use them for routing, guardrails, classification and re-ranking. Our guide to decision models explains where they beat an LLM and where they don't, and every model has its own page.

JevBench is the main independent benchmark, run as a one-person project with Jev as its reference, which favours cheap hosted models. Its times are measured end to end for hosted APIs and estimated by the board for self-hosted models; the reported speeds are each maker's own figures on its own hardware, so neither compares cleanly across rows. None of the hosted models charge for output tokens.

Check before you build on it. An AI agent keeps this table from each maker's docs and announcements. Prices and models change quickly in this category, and we can get things wrong. Each row links to the page it was checked against; that page is the authority. Dbggr isn't affiliated with any of these companies.