For model
providers.
Odin routes its own work across many models and brokers. The Odin R&D lab measures what that costs and how the providers behave, and publishes the results with their method and data. If you run models or route calls to them, you may be included when they fit an experiment's question. Providers in that experiment face its same pre-registered decision rules; nothing on this page promises inclusion or an outcome in advance.
Corrected on
Wording clarification · . The earlier lead said a provider “can be part of that measurement” and claimed universal provider rules; the earlier social preview repeated that claim. Inclusion depends on the question and terms of a particular run. The published records do not establish identical decision rules across all providers or automatic inclusion. Our publication standard explains this correction.
What we measure.
What a call costs as billed, not as estimated; whether the model that answers is the model that was asked for; how often calls fail; latency; and what a useful result costs, model by model. The results so far are on the provider-economics page, and the weekly broker probe repeats a small measurement every week.
Every experiment is pre-registered: its question, measure, decision rule and spend bound are published before any counted call, and a result that goes against us stays published. How we run experiments explains it, and selected evidence and open questions keeps what is measured apart from what is not.
What a provider gets.
- A place in a public measurement. When your models or routing fit an experiment's question, you may be included under its pre-registered decision rules. The weekly broker probe uses a separately published fixed method and reports observations without ranking brokers. Inclusion in either is decided for that run; it is not automatic.
- A conversation about real traffic. Odin routes its own work across providers. A partner can be considered for that routing; any volume would be agreed in writing, and none is promised on this page.
- A written case study. When an experiment includes you, a field note explains the method and the result, with its data, on this site.
- Known rules before the run. Each pre-registration is public before any counted call, so you know what will be measured and how it will be judged. Results are published whatever they show.
What we ask.
- Research credits. Token credits to run the experiments that include you. Credits never buy a result; the spend bound and the decision rule are fixed before the run.
- Pricing. Research or volume pricing we can use for Odin's own routing, stated in writing.
- Consent where your terms ask for it. We name a provider next to a measured number only when its published terms allow it, as recorded in
data/provider-terms.json. If yours require written consent, we ask for it; until then you are shown by letter.
Partners.
No partner agreement is signed yet.
A provider is listed here only after an agreement is signed and on record. We show no logos and no "in conversation with" lines.