GPT-5 launched in August 2025, and the ChatGPT model lineup has continued to change. OpenAI’s current help documentation notes that several earlier ChatGPT models—including the original GPT-5 Instant and Thinking variants—were retired in February 2026. API availability and ChatGPT subscription access are separate, so teams must check both before planning a migration.
Do not hard-code a production workflow around a marketing model name. Store the model identifier as configuration, record the prompt and expected output, and keep a small regression set of representative inputs. A newer model can change formatting, tool selection, latency, refusal behavior, and cost.
Before switching, compare factual accuracy, structured-output validity, tool-call reliability, editing time, latency, and total token cost on your own workload. Re-run privacy and data-residency checks because controls may differ between consumer workspaces, business plans, and APIs.
For automations built in Make, n8n, Zapier, or custom code, add explicit fallbacks for model errors and schema failures. Keep consequential actions behind an approval gate until the updated workflow passes a representative pilot.