Bezpośrednia odpowiedź
Ta strona wyjaśnia, jak zespoły korzystają z bramy NextModel zgodnej z OpenAI. How catalog ids are shaped, why they contain slashes, and which id to send where. Dodaje praktyczne kroki, uwagi konfiguracyjne i najczęstsze pytania.
The catalog id is the contract
Every model has exactly one public id, and GET /v1/models is the single source of truth for it. Ids are provider-prefixed with a slash: qwen/qwen3.7-max, anthropic/claude-sonnet-4.5, openai/gpt-image-1. The prefix names the vendor the same way the website's model pages do; it is part of the id and must be sent verbatim. A handful of early doubao chat models ship without a prefix (doubao-seed-2-0-pro and siblings) — those bare ids are still valid catalog ids.
Where each id goes
The same id travels in different fields depending on surface — the id itself never changes.
| API request body | The model field: "model": "qwen/qwen3.7-max" |
| Model list / retrieve | GET /v1/models or /v1/models/{id} — ids are matched against the raw path, so slashes need no URL encoding |
| Agent / tool config | Whatever field the tool calls model/model id/Model ID — paste the full catalog id |
| Website URL | Slashes become double dashes: qwen/qwen3.7-max → /models/qwen--qwen3.7-max |
When a model id fails
model_not_supported (400) means the id is not in the live catalog: check spelling against GET /v1/models, keep the provider prefix, and watch for lookalikes (dated builds such as qwen/qwen3-max-2026-01-23-glb are distinct ids from their stable siblings). Model names shown in marketing pages are display names, never request values.