OpenAI API Model Shutdown: IDs and Official Replacements

Your immediate task is to find these four model IDs in your code and configuration, then test their replacements. They are gpt-3.5-turbo-instruct, gpt-3.5-turbo-1106, babbage-002, and davinci-002.

OpenAI’s official table lists gpt-5.6-terra as the recommended replacement for all four models. “Official replacement model” in the title refers to this recommendation. It does not mean you must use this model exclusively.

The OpenAI API deprecation notice lists the four models in its September 26, 2025 entry. The shutdown date in the table is September 28, 2026 for all of them.

Where to look

Search your repository for the four IDs. Check environment variables and deployment configuration as well. Review batch jobs and external automation settings.

After finding each usage, validate the replacement model in a test environment. The official GPT-5.6 Terra model page explains its API availability and supported endpoints.

The recommended review sequence is as follows.

1. Find the services that use the affected IDs.
2. Define representative inputs and expected output formats.
3. Switch the model in a test environment.
4. Compare output content and structured formats.
5. Check error handling and response latency.
6. Review token usage and cost.
7. Prepare a rollback procedure, then deploy.

This article did not use live API calls, synthetic experiments, or hypothetical cases. The list above is a review procedure for readers to run themselves. It does not report verified quality improvements or cost reductions.

What the official documentation confirms

Item	Confirmed detail
Announcement entry date	September 26, 2025
Shutdown date in the table	September 28, 2026
Affected scope	Four model IDs
Recommended replacement model	gpt-5.6-terra

The official documentation does not state an exact shutdown time or regional differences. This article also did not call the four models directly to verify whether access is blocked. Check your current account-level access directly.

Where the impact may occur

The impact is limited to code and configuration that directly call the four IDs. This may include copy generation, classification, CRM automation, report generation, and SaaS features. Not every similarly named GPT-3.5 model is affected.

Managing model IDs in one place makes it easier to avoid omissions during the next migration. It is also useful to record the actual called model and version in deployment logs.

Operations checklist

• Current model ID
• Services that call it
• Representative test inputs
• Output formats that must be preserved
• Acceptable error and latency ranges
• How to review costs
• Deployment owner
• Rollback method

Principles to reuse for the next migration

• Decision principle: Determine the affected scope by the actual called ID, not the model family name.
• Operations principle: Validate output, errors, latency, and cost separately using representative inputs for the replacement model.
• Work template: Confirm model IDs → list services using them → fix test inputs → compare results → prepare rollback → review logs after deployment.

Frequently asked questions

Are all GPT-3.5 APIs affected?
No. The affected models confirmed in this announcement are the four IDs above.

What is the official replacement model?
The model officially recommended for all four IDs in the table is gpt-5.6-terra.

Can I simply change the model name?
No. You should recheck output formats, error handling, latency, token usage, and cost with the same inputs.

Official sources

• OpenAI API deprecation notice
• Official GPT-5.6 Terra model page

The official sources were checked again on September 29, 2026. The dates, model list, and recommended replacement model come from the official documentation. The review sequence is a practical recommendation.