Benchmarks & research
The vLLM Semantic Router team's Decision 2.0 is a set of six open decision models, from 0.6B to 27B parameters, each answering several typed questions about one message at once.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
The team says each size beats the same-size open models it compared on JevArena, though its 27B and 4B models only draw level with AutoJev-27B and Decider 4B. Its Decision Index numbers come from its own run of version 0.2.1, not the 0.3 named in the launch post. The weights are Apache-2.0, and loading any of them runs code that ships with the weights.
Use it when one message needs several answers at once. For a complaint about a damaged order, it can say which team should handle it, whether the customer has a receipt and how urgent it is, all in one go. The smallest model is the quickest; the largest is slower but can read much longer messages.