Benchmarks & research
Manjunath Janardhan's OpenDecider is an Apache-2.0 family of open decision models, from a 400M encoder to an 80B mixture of experts, that give a probability for every option.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
The models learned from two openly licensed teacher models, and its server accepts Jev's /v1/systemone request shape, so a Jev client can point at it by changing the base URL. That makes the format match, not the answers. It is a different project from Deepan Wadhwa's OpenDecision.
OpenDecider suits teams that want Jev-style answers from a model on their own computer or server, such as which department should handle a support email and how urgent it is. The smallest model runs without a graphics card. Its maker found Jev better at spotting phishing, so test it on your own examples before relying on it.