News · Model catalogue
Nine more models, and a prober that checks the rest
The catalogue now says which models odnoga can actually serve — and a conformance prober calls them, as a customer would, to keep that honest.
This piece was written by a model from odnoga’s own measured data. Every figure in it is checked against that data before publication.
odnoga's catalogue lists every model the platform can reach. Until this week it also listed 23 it could not.
What was wrong
Three separate causes, all of them structural rather than accidental:
- Video had no route at all. A request to a Veo model returned 404 before a vendor was ever chosen.
- Image and audio had adapters for OpenAI only. Google and xAI models returned
400 unsupported_vendor. - Realtime models speak a WebSocket protocol the gateway does not expose.
Each of those models had a public page telling you to call it.
What changed
Google image, speech and embeddings now work, and so does xAI image. Every request shape was confirmed against the live vendor API before a line of adapter was written — not taken from documentation.
Two additions to the catalogue: Gemini 3.8 Flash and Gemini Omni 1.1 Flash. Gemini 3.8 Flash is on promotional pricing until the end of 2026, so it carries two dated price rows rather than one. Your invoice will reflect the change on the day it happens, not a month later.
What keeps it honest
The fix that matters is not the adapters. It is that the catalogue now derives "can odnoga serve this?" from the serving code itself, and a conformance prober calls each model on its real endpoint — holding an ordinary API key, over the public URL, exactly as you would.
A model that stops answering stops being advertised. That is the point.