- Ya
- 29 Agosti 2026, 23:54 UTC
- Mwandishi
- Kamo
- Ahadi ya
- 618bf4c
Chat goes to the Inference Providers router **************** but discoverModels asked the HUB for models that are "warm" somewhere — a different and much larger set than the router will serve for a given token and enabled providers. So the model picker offered models the router then refused, and every message came back as 400 The requested model **************** is not supported by any provider you have enabled with nothing on screen to say which of the listed models were real. That id is one of two enabled for a live org and is not in the router's catalogue at all; 'deepseek-ai/DeepSeek-V3.2', enabled beside it, is. Discovery now reads {baseUrl}/v1/models — the router's own catalogue, which lists exactly what can be called. Models with no LIVE provider are dropped, because a model whose providers are all down answers the same 400 as one that was never routable and a member cannot tell them apart from a dropdown. Context length and tool support come off the best live provider; vision comes off the declared input modalities, instead of being hardcoded false. Also folds the duplicated base-url normalization into one helper and corrects the class javadoc, which still described the retired **************** shape.