OpenAI sets January 14 shutdown for selected GPT-4o and GPT-4.1 API models

OpenAI’s October 14, 2026 deprecation notice gives applications using selected GPT-4 models until January 14, 2027 to switch. It recommends GPT-6 Luna as their replacement. Migration guidance offers a concrete starting configuration: set reasoning effort to none, then test the model on representative workloads before moving production traffic.

The deadline follows the model version

Deprecation is the warning stage, not the moment access disappears. OpenAI says a model becomes deprecated when retirement is announced; it becomes inaccessible at shutdown. This notice covers specific dated releases, or snapshots, accessed through the API—the interface software uses to request model responses.

• GPT-4o: gpt-4o-2024-08-06 and gpt-4o-2024-11-20.
• GPT-4o mini: gpt-4o-mini-2024-07-18.
• GPT-4.1: gpt-4.1-2025-04-14.
• GPT-4.1 mini: gpt-4.1-mini-2025-04-14.

The dates leave three months between notification and shutdown. OpenAI’s general policy promises at least six months for generally available models, unless safety or compliance concerns require less. The October 14 entry does not explain how its interval fits that policy.

Check behavior before switching traffic

OpenAI expects most GPT-4.x use cases to migrate with the initial reasoning setting. It advises testing other settings to balance output quality, response time and cost, rather than assuming more reasoning is the right default.

The first step is to record the existing model’s quality, latency—the time taken to respond—and cost on representative requests, including important edge cases. Compare Luna against that baseline. Check output formatting and tool use where relevant, then adjust prompts, reasoning effort and verbosity and retest.

A cheaper token does not settle the bill

OpenAI says Luna has a lower per-token price than the GPT-4.x models covered by the migration and is more capable for most use cases. Tokens are the units used to meter model usage. Token consumption and reasoning settings can affect each task’s cost, so the company advises comparing total workload costs.

After testing, OpenAI recommends starting with a limited share of production traffic and monitoring quality, latency, errors and cost before completing the switch. Self-serve customers can seek help through help-center chat; sales-managed customers can contact their account teams to arrange engineering support and a product-specific migration plan.