OpenAI’s official deprecations page has two dated shutdown waves coming: Sep 28, 2026 (4 legacy models incl. gpt-3.5-turbo-instruct) and Oct 23, 2026 — the big one: 29 models including plain gpt-4, gpt-3.5-turbo, gpt-4-turbo, o1, o3-mini. After the date, calls return errors and any workflow naming those models stops producing output.
Why n8n workflows are extra exposed: the model id gets typed into an OpenAI node once — often years ago — and nobody looks again. HTTP Request nodes calling api.openai.com directly don’t show up when you skim credentials either.
5-minute audit: (1) export each workflow as JSON; (2) search the file for gpt-3.5, gpt-4, o1, o3-mini, o4-mini; (3) swap to the replacements OpenAI lists (gpt-5.6 family), run once end-to-end, and diff the output — the new models change tone/length/JSON adherence, so test before the deadline.
Disclosure: I run apibreaks.com, a deprecation-watch service for automation agencies. Full model→replacement table (sourced from OpenAI’s page, no signup): OpenAI shuts down gpt-4, gpt-3.5-turbo, o1 and 30+ legacy models — Sep 28 & Oct 23, 2026