OpenAI API Shutdown Dates 2026: Complete List and What to Do
The OpenAI API shutdown dates for 2026 span from August 2026 through January 2027, and the most significant one is October 23, 2026, when the entire GPT-4 era retires — including gpt-4, gpt-4-turbo, gpt-3.5-turbo, o1, and o4-mini. Before that, the Assistants API shuts down on August 26, 2026, and the Videos API and Sora 2 models shut down on September 24, 2026. If your application pins any of these model IDs, requests will fail with an error on the shutdown date — there is no grace period and no automatic fallback to a newer model.
This guide lists every announced shutdown date, the official replacement for each model, and the specific steps to take before each deadline.
How We Verified These Dates
Every date and replacement model in this guide was checked directly against OpenAI’s official API deprecations documentation on July 31, 2026, rather than compiled from secondary reporting. Where third-party sources conflicted with the official page — and several did, particularly around the July 2026 shutdown wave — we deferred to OpenAI’s own documentation.
We have flagged two things explicitly throughout: where OpenAI lists no replacement model (which changes your options entirely), and where a listed replacement is not a drop-in swap (which changes your timeline). Deprecation guides that omit these details tend to understate how much work a migration actually involves.
One caveat we want to state upfront rather than bury: OpenAI maintains its deprecations page live and can update recommended replacements after an initial announcement. Treat the tables below as accurate for the publication date and re-confirm the official page before you cut over.
What Does “Shutdown” Actually Mean?
OpenAI uses three terms that are easy to confuse, and the difference matters for planning.
Deprecated means OpenAI has announced the model is being retired. The model still works. You can keep calling it normally.
Shut down (also called “sunset”) is the date the model stops working entirely. A request that succeeded the day before returns an error the day after.
Legacy means the model no longer receives updates but has no announced shutdown date yet. Treat it as a signal to start planning a migration.
The practical takeaway: deprecation is a warning, shutdown is the deadline. Only the shutdown date breaks your code.
How Much Notice Does OpenAI Give Before a Shutdown?
OpenAI publishes minimum notice periods, and they vary significantly by model type:
- Generally available models: at least 6 months
- Specialized variants (chat variants like gpt-5.1-chat-latest, Codex variants like gpt-5.3-codex, deep research variants like o3-deep-research): at least 3 months
- Preview models (anything with preview in the name, such as computer-use-preview): as little as 2 weeks
That last category is the one that catches teams off guard. OpenAI explicitly advises against using preview models for business-critical production workloads unless you can migrate on short notice.
OpenAI notifies affected customers by email and documents every deprecation on its official deprecations page. If safety or compliance concerns force a faster timeline, notice periods can be shorter than the minimums above.
Complete List of OpenAI API Shutdown Dates in 2026 and 2027
August 10, 2026 — Chat-latest snapshots
| Model | Replacement |
| gpt-5.2-chat-latest | gpt-5.6-sol |
| gpt-5.3-chat-latest | gpt-5.6-sol |
Announced May 8, 2026. This is under two weeks away at the time of writing — check your configs now.
August 26, 2026 — Assistants API
| System | Replacement |
| Assistants API | Responses API and Conversations API |
Announced a full year in advance, on August 26, 2025. This is a complete product discontinuation rather than a model swap, so migration requires real code changes, not a string replacement. OpenAI publishes an Assistants-to-Conversations migration guide for this transition.
September 24, 2026 — Videos API and Sora 2
| Model / System | Replacement |
| Videos API | None listed |
| sora-2 | None listed |
| sora-2-pro | None listed |
| sora-2-2025-10-06 | None listed |
| sora-2-2025-12-08 | None listed |
| sora-2-pro-2025-10-06 | None listed |
Note the important detail: OpenAI lists no recommended replacement for the Videos API or any Sora 2 model. Unlike every other entry on this page, there is no in-platform migration target. If your product depends on OpenAI video generation, you need to evaluate alternative providers rather than plan a swap — and that is a longer project than a model migration.
September 28, 2026 — Older completions models
| Model | Replacement |
| gpt-3.5-turbo-instruct | gpt-5.4-mini or gpt-5-mini |
| babbage-002 | gpt-5.4-mini or gpt-5-mini |
| davinci-002 | gpt-5.4-mini or gpt-5-mini |
| gpt-3.5-turbo-1106 | gpt-5.4-mini or gpt-5-mini |
October 23, 2026 — The GPT-4 era ends
This is the largest single shutdown on the calendar.
| Model | Replacement |
| gpt-3.5-turbo-0125, gpt-3.5-turbo | gpt-5.6-terra |
| gpt-4-0613, gpt-4 | gpt-5.6-sol |
| gpt-4-1106-preview | gpt-5.6-sol |
| gpt-4-turbo, gpt-4-turbo-2024-04-09 | gpt-5.6-sol |
| gpt-4.1-nano, gpt-4.1-nano-2025-04-14 | gpt-5.6-luna |
| gpt-4o-2024-05-13 | gpt-5.6-sol |
| gpt-image-1 | gpt-image-2 |
| o1-2024-12-17, o1 | gpt-5.6-sol |
| o1-pro-2025-03-19, o1-pro | gpt-5.6-sol (reasoning.mode: pro) |
| o3-mini-2025-01-31, o3-mini | gpt-5.6-sol |
| o4-mini-2025-04-16, o4-mini | gpt-5.6-terra |
| ft-o4-mini-2025-04-16 | gpt-5.6-terra |
Fine-tuned models also retire on this date:
| Model | Replacement base model |
| ft-gpt-3.5-turbo | gpt-5.4-mini |
| ft-gpt-4 | gpt-5.5 |
| ft-gpt-4.1-nano-2025-04-14 | gpt-5.4-nano |
| ft-babbage-002 | gpt-5.4-mini |
| ft-davinci-002 | gpt-5.4-mini |
If you have custom fine-tuned models built on these bases, you will need to retrain on a new base model. That means data preparation, training runs, and evaluation — a genuine project rather than a config change. Budget time accordingly.
October 31 and November 30, 2026 — Platform products
Three platform products announced on June 3, 2026 are shutting down, not just models:
| Date | What happens |
| Oct 31, 2026 | Existing evals become read-only |
| Nov 30, 2026 | Evals dashboard and API shut down |
| Nov 30, 2026 | v1/prompts API and reusable prompt objects shut down |
| Nov 30, 2026 | Agent Builder shuts down |
For reusable prompts, the migration path is to move prompt content into your application code. For Agent Builder, OpenAI points users to the Agents SDK or ChatGPT Workspace Agents. ChatKit remains available and is not affected by this wave.
December 1, 2026 — GPT Image models
| Model | Replacement |
| gpt-image-1-mini | gpt-image-2 |
| gpt-image-1.5 | gpt-image-2 |
| chatgpt-image-latest | gpt-image-2 |
Worth noting: developers who migrated from DALL·E 2 or DALL·E 3 to gpt-image-1-mini when those retired on May 12, 2026 now face a second forced migration within roughly six months. OpenAI is consolidating all image generation onto gpt-image-2.
December 11, 2026 — GPT-5 and o3 snapshots
| Model | Replacement |
| gpt-5-2025-08-07 | gpt-5.6-sol |
| gpt-5-mini-2025-08-07 | gpt-5.6-terra |
| gpt-5-nano-2025-08-07 | gpt-5.6-luna |
| gpt-5-pro-2025-10-06 | gpt-5.6-sol (reasoning.mode: pro) |
| o3-2025-04-16 | gpt-5.6-sol |
| o3-pro-2025-06-10 | gpt-5.6-sol (reasoning.mode: pro) |
This is the one most likely to surprise teams, because these are not old models. If you pinned an exact dated snapshot ID for reproducibility — which is normally good practice — you are affected.
January 6, 2027 — Fine-tuning job creation ends
Active existing customers can no longer create new fine-tuning jobs after this date. Inference on already fine-tuned models continues until the underlying base model is deprecated.
Two earlier restrictions already took effect: from May 7, 2026, organizations that had never run fine-tuning could no longer create jobs, and from July 2, 2026, organizations that had not run inference on a fine-tuned model in the past 60 days lost access to job creation.
January 20, 2027 — Legacy audio, realtime, and transcription
| Model | Replacement |
| gpt-realtime | gpt-realtime-2.1 |
| gpt-audio | gpt-audio-1.5 |
| gpt-4o-audio | gpt-audio-1.5 |
| gpt-4o-realtime | gpt-realtime-2.1 |
| gpt-realtime-mini | gpt-realtime-2.1-mini |
| gpt-audio-mini | gpt-audio-1.5 |
| gpt-4o-mini-realtime | gpt-realtime-2.1-mini |
| gpt-4o-mini-audio | gpt-audio-1.5 |
| gpt-4o-mini-transcribe-2025-03-20 | gpt-4o-mini-transcribe-2025-12-15 |
Announced July 20, 2026.
Which OpenAI API Shutdown Dates Have Already Passed?
If your application broke recently and you are trying to work out why, this is the likely cause. Fifteen model snapshots shut down on July 23, 2026, announced back on April 22:
- Codex lineage: gpt-5-codex, gpt-5.1-codex, gpt-5.1-codex-max, gpt-5.1-codex-mini, gpt-5.2-codex
- Chat aliases: gpt-5-chat-latest, gpt-5.1-chat-latest
- Deep research: o3-deep-research-2025-06-26, o4-mini-deep-research-2025-06-26
- Computer use: computer-use-preview-2025-03-11
- Search preview: gpt-4o-search-preview-2025-03-11, gpt-4o-mini-search-preview-2025-03-11
- Audio and TTS: gpt-4o-mini-tts-2025-03-20, gpt-audio-mini-2025-10-06, gpt-realtime-mini-2025-10-06
Most of these point to gpt-5.6-sol or gpt-5.6-terra as replacements. Earlier in 2026, DALL·E 2 and DALL·E 3 retired on May 12, and a batch of legacy GPT-4 snapshots retired on March 26.
What Are GPT-5.6 Sol, Terra, and Luna?
Almost every replacement in the tables above points to a model named Sol, Terra, or Luna. OpenAI dropped size-based naming (mini, nano, pro) in the GPT-5.6 generation and replaced it with these names:
- gpt-5.6-sol — the flagship model, and the replacement for gpt-4, gpt-4-turbo, gpt-4o, o1, o3, and gpt-5
- gpt-5.6-terra — the mid-tier model, replacing gpt-3.5-turbo, o4-mini, and gpt-5-mini
- gpt-5.6-luna — the lightweight model, replacing gpt-4.1-nano and gpt-5-nano
For workloads that previously used o1-pro, o3-pro, or gpt-5-pro, the replacement is gpt-5.6-sol with the reasoning.mode: pro parameter rather than a separate model ID. This is a structural change worth noting: if your code branches on model name to select a reasoning tier, that logic needs rewriting, not just renaming.
What Should You Do Before Each Deadline?
Audit what you actually call. Search your codebase, environment variables, and config files for hardcoded model IDs. Include background jobs, cron tasks, and internal tools — these are consistently the ones teams forget. Pinned dated snapshots (anything ending in a date like -2025-08-07) are the highest risk.
Check whether you use aliases or snapshots. An alias like gpt-4 and a snapshot like gpt-4-0613 can behave differently, and in this deprecation wave OpenAI grouped both under the same October 23 date.
Do not assume replacements are drop-in. OpenAI’s recommended replacement is a starting point, not a guarantee of identical behaviour. Newer models may handle your prompts differently, return different output formats, or support different parameters. Test before you cut over, particularly for structured output and function calling.
Migrate in staging first, using real traffic samples. Run your actual production prompts through the replacement model and compare outputs side by side. Quality regressions are far easier to catch before a deadline than during an outage.
Budget separately for fine-tuned models. These require retraining on a new base model. Treat October 23 as a project deadline, not a checkbox.
Re-verify dates before you cut over. OpenAI maintains its deprecations page live and can update recommended replacements after an announcement. Always confirm the current table on the official page rather than relying on any blog post — including this one.
Track platform-by-platform. These dates apply to the direct OpenAI API. Azure OpenAI and Amazon Bedrock run separate model lifecycles with their own retirement calendars. If you run models on more than one platform, track each schedule independently.
Can You Keep Using a Model After Its Shutdown Date?
In some cases, yes. OpenAI notes that developers may be able to provision dedicated capacity for continued access after a model’s shutdown date, and directs teams to contact its sales team to explore that option. This is not self-serve and is unlikely to be practical for smaller projects, but it exists for organisations with a genuine constraint.
For everyone else, the shutdown date is final.
Frequently Asked Questions About OpenAI API Shutdown Dates
What are the most important OpenAI API shutdown dates in 2026? October 23, 2026 is the most significant, retiring the largest group of models in a single day, including gpt-4, gpt-4-turbo, gpt-3.5-turbo, gpt-4o-2024-05-13, o1, o1-pro, o3-mini, o4-mini, gpt-4.1-nano, gpt-image-1, and five fine-tuned model families. August 26 (Assistants API) and December 11 (GPT-5 and o3 snapshots) are the next most impactful.
What happens if I do not migrate before the shutdown date? Requests targeting the retired model ID stop being served and return an error instead of a completion. There is no automatic fallback to a newer model and no grace period built into the model IDs.
Is GPT-4 still available in the API right now? Yes. As of July 2026 the gpt-4 API entry is still listed and functional. It shuts down on October 23, 2026. Note that GPT-4 was removed from the ChatGPT product interface much earlier, on April 30, 2025 — the API and the consumer product follow separate timelines.
Do these shutdown dates apply to Azure OpenAI? No. Azure OpenAI runs a separate model lifecycle with its own retirement notices and dates, as does Amazon Bedrock. If you deploy through those platforms, track their schedules independently.
What replaces the Assistants API? The Responses API together with the Conversations API. OpenAI published a migration guide for this transition. The Assistants API shuts down on August 26, 2026.
Is there a replacement for Sora 2 and the Videos API? OpenAI lists no recommended replacement for either. Both shut down on September 24, 2026, so teams relying on OpenAI video generation should evaluate alternative providers rather than plan an in-platform migration.
How much notice does OpenAI give before retiring a model? At least 6 months for generally available models, at least 3 months for specialized variants such as chat, Codex, and deep research models, and as little as 2 weeks for preview models.
Where can I check the official OpenAI API shutdown dates? OpenAI maintains its deprecations page in its API documentation, updated live. Since recommended replacements can change after an announcement, always confirm the current table there before migrating.
