OpenAI Codex & GPT-5 Snapshots Retired (2026-07-23)

Last verified: 2026-07-16 — checked against the OpenAI API Deprecations page. This page covers the 2026-07-23 retirement batch, which has already taken effect.

Direct answer

On 2026-07-23, OpenAI retired the Codex snapshots (gpt-5-codex, gpt-5.1-codex, gpt-5.1-codex-max, gpt-5.1-codex-mini, gpt-5.2-codex), the chat-latest aliases (gpt-5-chat-latest, gpt-5.1-chat-latest), two gpt-4o search previews, one gpt-4o-mini-tts snapshot, and the March 2025 computer-use-preview. What to do if you still pin them: find every place these exact model IDs are pinned and repoint them to the official replacements in the table below (mostly gpt-5.5 and gpt-5.4-mini), then re-run your evaluations — the replacements are newer models, not byte-for-byte drop-ins.

Who this affects

Any API team or Codex user who still has one of these model IDs pinned in production — in application code, environment variables, CI config, SDK defaults, or an internal routing layer. Since 2026-07-23, requests that reference a retired ID fail rather than silently fall back.

Retirement table (2026-07-23)

Every date and replacement below is taken from the OpenAI API Deprecations page for the 2026-07-23 batch.

Retired model (shut down 2026-07-23)	Official replacement	Notes
gpt-5-codex	gpt-5.5	General model; re-test coding prompts
gpt-5.1-codex	gpt-5.5	General model; re-test coding prompts
gpt-5.1-codex-max	gpt-5.5	General model; re-test coding prompts
gpt-5.1-codex-mini	gpt-5.4-mini	Mini tier
gpt-5.2-codex	gpt-5.5	General model; re-test coding prompts
gpt-5-chat-latest	gpt-5.5	Rolling chat-latest alias retired
gpt-5.1-chat-latest	gpt-5.5	Rolling chat-latest alias retired
gpt-4o-search-preview-2025-03-11	gpt-5.4-mini	Not a drop-in — see caveats
gpt-4o-mini-search-preview-2025-03-11	gpt-5.4-mini	Not a drop-in — see caveats
gpt-4o-mini-tts-2025-03-20	gpt-4o-mini-tts-2025-12-15	Dated snapshot bump within the same TTS family
computer-use-preview-2025-03-11	computer-use-preview (or gpt-5.4-mini)	Move to the undated alias, or the general model

Replacements that are not drop-in

Three rows need more than a string swap:

• Search previews → gpt-5.4-mini. The *-search-preview models exposed built-in web search as a dedicated model. The replacement is a general model, so web search moves to the web search tool on the Responses API rather than being implied by the model name. Rework the request shape, not just the ID.
• computer-use-preview-2025-03-11 → computer-use-preview. The undated alias continues; pinning the dated March 2025 snapshot is what breaks. If you no longer need computer-use, gpt-5.4-mini is offered as the general alternative.
• gpt-4o-mini-tts snapshot bump. Same TTS family, newer dated snapshot (2025-12-15). Confirm voice/format parameters are unchanged for your usage.

Migration steps

1. Inventory every pinned ID. Grep your code, env files, and config for each string in the table (for example gpt-5-codex, gpt-5.1-codex-max, gpt-5-chat-latest, search-preview, computer-use-preview-2025-03-11). Include SDK defaults and infrastructure-as-code.
2. Map each ID to its replacement using the table above. Flag the three non-drop-in rows for extra work.
3. Re-run your evaluations against the replacement. gpt-5.5 and gpt-5.4-mini are different models — output style, tool-calling behavior, context limits, and pricing may differ from the retired snapshot. Verify prompts, structured-output schemas, and token budgets before cutover.
4. Roll out behind a canary. Send a slice of traffic to the replacement, compare against your quality and latency baselines, then ramp to full traffic.
5. Centralize the model ID so the next forced migration is one edit, not N. If your model IDs live in application code in many places, a mandatory retirement like this one means touching every call site. Routing through an OpenAI-compatible gateway turns it into a single config change. As an engineering illustration, byesu is an OpenAI-compatible API gateway where one key reaches current OpenAI models (gpt-6.1-sol, the gpt-6 family, gpt-5.6) and models from other providers behind the standard OpenAI request shape, so the model mapping is maintained in the routing layer instead of scattered across services. A gateway only serves the models it lists — byesu does not currently offer gpt-5.5 or gpt-5.4-mini, so check its model list before choosing a target. Third-party gateways bill at their own published rates — check the provider's own price list rather than assuming parity with OpenAI.

Verify before you cut over

Do not assume behavioral parity. Treat each replacement as a new model and run your own evals before moving production traffic; the retired IDs no longer respond, so compare against your stored baselines rather than live calls.

Heads-up: later retirement batches

The 2026-07-23 batch is not the last one. OpenAI's API Deprecations page also lists later retirement batches covering additional older GPT-5 and o-series snapshots. The exact model IDs, replacement models, and effective dates for those batches can still change, so this page does not pin them. If you rely on older GPT-5 or o-series snapshots, check the official page for the next batch that affects you and plan that migration ahead of its date.

Source

• OpenAI API Deprecations — https://developers.openai.com/api/docs/deprecations

Always confirm dates and replacement IDs against the official page above before you migrate; deprecation schedules can change.

FAQ

Which OpenAI models were shut down on 2026-07-23?

gpt-5-codex, gpt-5.1-codex, gpt-5.1-codex-max, gpt-5.1-codex-mini, gpt-5.2-codex, gpt-5-chat-latest, gpt-5.1-chat-latest, gpt-4o-search-preview-2025-03-11, gpt-4o-mini-search-preview-2025-03-11, gpt-4o-mini-tts-2025-03-20, and computer-use-preview-2025-03-11. Since that date, requests to these exact model IDs are no longer served.

What replaces gpt-5-codex and the other Codex snapshots?

OpenAI pointed gpt-5-codex, gpt-5.1-codex, gpt-5.1-codex-max, and gpt-5.2-codex to gpt-5.5, and gpt-5.1-codex-mini to gpt-5.4-mini. These are newer general and mini models, not identical drop-ins, so re-run your evaluations before switching production traffic.

What happens if my code still uses a retired model ID?

Since the 2026-07-23 shutdown, API calls that pin the retired model IDs are no longer served and return an error instead of a completion. Any integration that still references these IDs stays broken until it is updated to a supported model.

Are there more OpenAI model retirements after 2026-07-23?

Yes. OpenAI's API Deprecations page lists one or more additional retirement batches after the 2026-07-23 wave, covering older GPT-5 and o-series snapshots. The specific model IDs, replacement models, and effective dates for those later batches can still change, so confirm them on the official OpenAI API Deprecations page closer to the time rather than relying on a fixed list here.