OpenAI retires GPT-4o family after three months notice

Builders must audit every GPT-4.1 or GPT-4o endpoint for reasoning_effort, Responses API usage and temperature settings. An illustrative scenario: a coding agent that previously called GPT-4.1 with temperature 0.2 and function calling will now need separate non-reasoning calls or migration to Responses; the decision criteria are whether the project can absorb the API change before January and whether dedicated capacity can be arranged with OpenAI sales. Unknowns include exact parity of 1M context handling and diff reliability on large files.

Operators face a compressed migration that risks production downtime for any service still routing traffic to the retiring models. An illustrative scenario: a customer-support automation platform using fine-tuned GPT-4o snapshots must either schedule dedicated capacity before the deadline or accept the Luna behavior changes; eligibility requires checking the List models API shutdown_date field and confirming no fine-tuning is needed after 6 January. Data-handling terms remain unchanged but the short window limits testing time.