tazo-large is the default when model is omitted.
Check capabilities at runtime
CallGET /v1/models to see each model’s
capabilities and whether it is ready. Use this instead of hard-coding limits.
Legacy model names
Older internal model IDs and the previousgeneral and specialised aliases
are still accepted as input during migration. Responses and /v1/models always
use tazo-large and tazo-small.
If you compare the request model with the response model, switch your
requests to the new names. Aliases are resolved before idempotency checks, so a
retry that uses an alias or its legacy ID reuses the same job.