開發者平台
OpenAI Shuts Down `gpt-5.2-chat-latest`; Migration Requires More Than Replacing the Model Name
OpenAI discontinued `gpt-5.2-chat-latest` in the API on August 10 and officially recommends migrating to GPT‑5.5. The retirement targets an alias that tracks ChatGPT behavior, so engineering teams must still revalidate prompts, tool calls, and output formats rather than treating the migration as an equivalent routing switch.
OpenAI scheduled the API model alias `gpt-5.2-chat-latest` to be shut down on August 10, ending the migration period that began with its May notice. The recommended replacement is GPT‑5.5. The affected model is an alias presented as the “latest ChatGPT version.” It does not carry the same lifecycle commitment as the dated GPT‑5.2 checkpoints still listed in the documentation. Operations teams should therefore inventory the model IDs actually sent by each application instead of drawing conclusions from names shown in the console or software packages.
The technical risk extends beyond the endpoint beginning to return `model-not-found`. Applications often use `latest` aliases to obtain a response style similar to ChatGPT’s. Switching directly to GPT‑5.5 may change adherence to system prompts, JSON and structured output, tool selection, refusal boundaries, token usage, and latency. If an agent treats the old model’s natural-language tool arguments, reasoning text, or particular stop sequences as an implicit protocol, upgrading the model may also cause behavioral regressions that do not immediately produce errors.
A safer migration approach is to preserve a representative set of requests and outputs from the old model, then run offline replays against the new model. Evaluation should go beyond subjective quality comparisons to measure schema pass rates, tool-call success rates, retry counts, P95 latency, and per-task cost. Cache keys, tracing data, and evaluation reports should also record the full model identifier so that results from before and after the upgrade are not mixed together.
The documentation currently specifies only the retirement of `gpt-5.2-chat-latest`; this should not be taken to mean that every GPT‑5.2 snapshot will disappear on the same day. Engineering teams should now watch for whether OpenAI announces separate retirement dates for pinned snapshots. They should also verify whether SDKs, model gateways, and third-party agent frameworks explicitly fail, silently fall back, or automatically route requests to another model when they encounter a removed alias.