The plugin AI-model catalog now lists each model under only the providers that can actually run it. Until today, GET /api/v1/plugin/ai-models and GET /api/v1/plugin/pricing returned gpt-image-2 twice: the correct openai/gpt-image-2 entry at 60 credits, and a second byteplus/gpt-image-2 entry at 10 credits left over from a placeholder used while the OpenAI integration was still in development.
That duplicate was not just visual clutter. Because the byteplus provider passes the runtime availability check, a plugin could submit a job against byteplus/gpt-image-2, have only 10 credits reserved, and then watch the job fail in the backend — which has no GPT Image model on the BytePlus provider. The catalog now derives a single, correct entry per model: gpt-image-2 is offered exclusively as openai/gpt-image-2 at 60 credits, and the ghost byteplus/gpt-image-2 is gone from both /ai-models and /pricing. No action is required from integrators who already used the openai/gpt-image-2 identifier.