loading…
Market prices/compare
Each page below prices one model against another from the live book: best asks, the cost of a real request, what settled requests paid, router-measured latency and throughput, and the observed task-fit profile of each class. Nothing is interpolated, and any figure the market has not produced reads n/a.
28 model classes, 378 head-to-head pages. The set is the top of the reviewed independent quality ranking, plus the leading model of each vendor family, so a cross-vendor question always has a page.
One block per model, listing every model it can be compared with. The price beside a heading is that model's best live output ask per 1M tokens.
Omnious has no price list. Providers quote input and output rates per million tokens in USDC, and every request is an RFQ: the quotes competing for that model class enter a live scoring auction that weighs quoted price against router-measured latency and reliability.
The winner clears under second-score/3 rules. It is paid a price capped by the next genuine rival's score, so quoting low to win and billing high is structurally impossible. Settlement is pay-as-you-go USDC on the actual token counts of the response, and every response carries a signed, verifiable receipt.
Every figure on these pages is therefore a reading of the market at a point in time: the best asks on the book, and the average cleared prices of requests that settled. They move whenever the market does.
One OpenAI-compatible key reaches every model on this page. Switch between them per request and settle in USDC on actual usage.
Get an API key