Awesome Decision Models

No, and the list says so up front: scores are as reported by each project on the benchmark that project chose, so numbers are not comparable across entries. It is a list, not a benchmark. Follow any figure back to its own project before treating it as measured.

No. It covers hosted APIs, open-weight models, inference techniques, runtimes and platforms, SDKs and clients, applications, demos, benchmarks, papers and articles, including runtimes that serve /v1/systemone without TypeSafe, such as jevos, jev-style, Kev and Von. It is community-maintained, not affiliated with any model provider, and published in English and Chinese with a site mirror.

It varies by entry and the list flags the split cases: OpenJev's weights are CC BY-NC 4.0 for noncommercial use while its helper and server code is Apache-2.0, whereas entries such as Clef, GLiNER2.5-Decide and Standard One are Apache-2.0. Read each project's own terms before shipping.

Some. jevos is the CPU-oriented entry: a 1B model cut to 17 layers, GGUF q4_k_m at 619 MB, running on llama.cpp at about 54 ms per short request. Its limit is scope, not hardware: it speaks the wire format for Noul questions only, and Choice and Score return 422.

A curated list maintained by AnotiaWang, community-maintained and not affiliated with any model provider, with a website at anotiawang.github.io/awesome-decision-models and an English and simplified Chinese README.

Hosted APIs, open models from companies and the community, inference techniques, runtimes and platforms, SDKs and clients, applications, demos and games, agent tools, benchmarks and evaluations, papers, articles, and related resources.

API-only models are listed under Hosted APIs, while open-weight models their publishers also host, such as Clef and pplx-decider, appear under Open Models, where each entry notes its license and where it runs.