OpenAI will shut down a group of older API model aliases and snapshots on July 23, 2026. The affected list spans Codex, Search, Computer Use, text-to-speech, Audio, Realtime, Chat, and Deep Research. Applications that still send one of these IDs may fail after the deadline, so teams should finish discovery and regression testing before changing production routing.
Action deadline: July 23, 2026
Do not wait for the shutdown to reveal hidden dependencies. Search code, configuration, secrets, scheduled jobs, database records, and runtime logs now.
Official replacement table
| Retiring model or snapshot | OpenAI recommended replacement |
|---|---|
computer-use-preview-2025-03-11 / computer-use-preview | gpt-5.4-mini |
gpt-4o-mini-search-preview-2025-03-11 | gpt-5.4-mini |
gpt-4o-mini-tts-2025-03-20 | gpt-4o-mini-tts-2025-12-15 |
gpt-4o-search-preview-2025-03-11 | gpt-5.4-mini |
gpt-5-chat-latest | gpt-5.5 |
gpt-5-codex | gpt-5.5 |
gpt-5.1-chat-latest | gpt-5.5 |
gpt-5.1-codex | gpt-5.5 |
gpt-5.1-codex-max | gpt-5.5 |
gpt-5.1-codex-mini | gpt-5.4-mini |
gpt-audio-mini-2025-10-06 | gpt-audio-1.5 |
gpt-realtime-mini-2025-10-06 | gpt-realtime-mini |
o3-deep-research-2025-06-26 / o3-deep-research | gpt-5.5-pro |
o4-mini-deep-research-2025-06-26 / o4-mini-deep-research | gpt-5.5-pro |
gpt-5.2-codex | gpt-5.5 |
Source: OpenAI API Deprecations. Check the official page again immediately before migration in case the schedule changes.
Step 1: Find every dependency
A repository search is necessary, but it is not sufficient. Model IDs commonly live outside the main application code. Build an inventory across:
- application source, SDK wrappers, tests, fixtures, and example configuration;
- local, preview, staging, and production environment variables;
- database rows used for model routing, feature flags, or tenant configuration;
- scheduled jobs, queues, serverless functions, and workflow automation;
- CI/CD secrets and deployment-platform settings;
- third-party agent builders and integration platforms;
- production request logs, which can expose dynamically supplied model IDs.
Example repository scan
rg -n 'computer-use-preview|gpt-4o-(mini-)?search-preview|gpt-5(\.1|\.2)?-(chat-latest|codex)|gpt-(audio|realtime)-mini|o[34].*deep-research' .Step 2: Treat each replacement as a migration
A recommended replacement is not a guarantee of identical behavior. Moving a Chat or Codex workload to gpt-5.5 can change reasoning behavior, latency, token consumption, and output style. Search and Computer Use migrations may require tool-level changes. Audio and Realtime applications need end-to-end session testing, not only a successful API response.
Quality checks
Run representative prompts, tool calls, structured-output tests, safety checks, and human review against the current and replacement models.
Operational checks
Compare latency, retries, rate limits, context usage, token cost, timeout behavior, observability, and rollback readiness.
Step 3: Roll out with a rollback path
- Create a complete inventory of affected workloads and owners.
- Run an evaluation set against the current and replacement models.
- Update prompts or tool schemas only where testing shows they are needed.
- Deploy behind a feature flag or model-routing configuration.
- Start with internal or low-risk traffic and monitor errors, quality, latency, and cost.
- Keep the previous configuration documented for rollback until validation is complete.
- Remove the old IDs before July 23 and repeat the search to confirm zero remaining dependencies.
AILinkBase recommendation
Prioritize production workloads that use Codex, Search, Computer Use, Audio, Realtime, or Deep Research. Do not change unrelated historical articles, benchmark records, or archived examples merely because they mention an older model; migration work should target active runtime dependencies first.
Related: GPT-5.6 model guide · GPT-5.6 and ChatGPT Work · Best AI tools for developers