As of June 12, GPT-5.2 Instant, GPT-5.2 Thinking, and GPT-5.2 Pro are no longer selectable in ChatGPT. OpenAI is automatically routing any open conversation that was running on a 5.2 model onto the corresponding GPT-5.5 variant. The retirement was pre-announced when GPT-5.3 Instant shipped, but the actual cutover landed without an extended phase-out period.

The mechanical detail matters: when conversation history carries over to a different model, output quality and tone shift even if the chat looks continuous. Developers relying on consistent behavior across long-running threads need to revalidate, and teams that pinned to 5.2 for cost or latency reasons are now on the newer family's price card whether they planned to be or not.

This is the pattern OpenAI has settled into across 2026 — frequent point-release pruning rather than long, parallel support windows. Each retirement narrows the surface area the company has to maintain and concentrates traffic on the latest checkpoints, which also happens to produce more training-relevant feedback. The trade-off lands on downstream teams that have to track an accelerating deprecation calendar.

Takeaway for learners: if you're building on a hosted model, the model is not a stable dependency. Treat the model ID like a runtime version that can be pulled at short notice, write evals you can run against any successor in minutes, and keep a written record of which version you tested against — "GPT-5.2 worked" is not a fact you can rely on a week later.