Streaming Tokens and Tools: Multi-Turn Agentic Harness Support in NVIDIA Dynamo | NVIDIA Technical Blog
… In this setup, both routes reached OpenAI-served GPT-5.5 — the difference was the endpoint and the model-catalog record Codex attached to it. When the custom endpoint used model ID openai/openai/gpt-5.5 without being associated with the gpt-5.5 catalog profile, Codex used generic fallback behavior. …