Sunsets and Deprecations: The Timeline to Follow
On this page
New features come with a major cleanup. Between July 2026 and April 2027, OpenAI is shutting down or retiring several products and a long list of models. Some deadlines are imminent: if your code or workflows depend on them, you need to act now. This lesson summarizes the official deprecations page and ChatGPT release notes as of November 5, 2026.
What’s Already Closed
The Assistants API shut down on August 26, 2026. Any integration still using it has stopped working; the documented migration path moves to the Responses API and the Conversations API. If you encounter a tutorial that creates “assistants,” “threads,” and “runs,” it describes this legacy API.
Other sunsets followed. On September 24, the Videos API and Sora 2 models left the API. On October 23, a series of older models were retired, including o1, o1-pro, o3-mini, o4-mini, GPT-4, GPT-4 Turbo, GPT-3.5 Turbo, and gpt-image-1. On the ChatGPT side, the Atlas browser stopped on August 9, o3 left ChatGPT on August 26, and GPT-5.5 on October 14.
Upcoming Deadlines
| Date | What’s Stopping | Recommended Replacement |
|---|---|---|
| November 30, 2026 | Agent Builder | Agents SDK or ChatGPT workspace agents |
| November 30, 2026 | Evals Platform (dashboard and API), read-only since October 31 | Promptfoo, per OpenAI’s migration guide |
| November 30, 2026 | Reusable Prompts (v1/prompts) | Embed prompt content directly in your application code |
| December 1, 2026 | gpt-image-1.5, gpt-image-1-mini, chatgpt-image-latest | gpt-image-2.6-sunburst or gpt-image-2.6-flare |
| December 11, 2026 | gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3, o3-pro | Corresponding GPT-5.6 models (Lesson 8) |
| December 11, 2026 | Custom GPTs in ChatGPT (February 11, 2027 for FedRAMP and Enterprise spaces with waiver) | Plugins (Lesson 11) |
| January 6, 2027 | tts-1, tts-1-hd, gpt-4o-mini-tts | gpt-realtime-2.1-mini |
| January 6, 2027 | Creation of new fine-tuning jobs for existing customers | — |
| January 14, 2027 | GPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini | gpt-6-luna |
| January 20, 2027 | Legacy audio and realtime models (gpt-realtime, gpt-audio, gpt-4o-realtime…) | gpt-realtime-2.1, gpt-realtime-2.1-mini, gpt-audio-1.5 |
| February 1, 2027 | gpt-5.5-cyber-preview | The most capable cybersecurity model available to you |
| February 26, 2027 | whisper-1, gpt-4o-transcribe (and its mini and diarize variants), /v1/audio/translations | gpt-transcribe or gpt-live-transcribe; v1/realtime/translations |
| April 1, 2027 | gpt-5.1, gpt-5.3-codex, gpt-5.4-nano | gpt-6-sol for the first two, gpt-6-luna for the third |
Fine-tuned versions of affected models follow their base model’s schedule.
The Fine-Tuning Situation
Self-service fine-tuning deserves a separate explanation because it isn’t disappearing all at once. Since May 7, 2026, organizations that had never performed fine-tuning can no longer start new jobs; since July 2, job creation has been closed to organizations that haven’t used a fine-tuned model in the last 60 days; on January 6, 2027, no customer will be able to launch a new job. Already trained models remain usable until their base model is retired. If your strategy relied on fine-tuning, pivot now toward prompt engineering, RAG, and evaluations.
Action Plan
Start with an inventory: search your code for model IDs, calls to assistants, evals, prompts, and fine_tuning, and audio model names. Next, sort results by deadline: anything touching November 30 and December 1 takes priority. For each replacement, run a new evaluation campaign on your own examples using a tool that will survive past November 30, since OpenAI’s Evals platform closes on that date.
Finally, add the deprecations page to your monthly review cycle. OpenAI commits to minimum notice periods: at least six months for a generally available model, at least three months for a specialized variant like a Codex version or a deep research version, and sometimes only two weeks for a preview model, whose name contains preview. Affected customers receive email notifications, but regular reviews prevent surprises.
Your Turn
Build your timeline. Revisit the inventory started in Lesson 1 and the research from Lesson 8. For each affected dependency, note the sunset date, the recommended replacement, the person responsible, and the date you will finish testing. Prioritize everything stopping on November 30 and December 1, 2026.
Official Sources: Deprecations · Changelog (OpenAI API documentation) · ChatGPT release notes · Custom GPT retirement and migration FAQ (OpenAI Help Center)
Key Takeaways
- The Assistants API closed on August 26, 2026;
o4-mini,o1, ando3-minion October 23 - On November 30, 2026, Agent Builder, the Evals platform, and reusable prompts close
- Legacy image models stop on December 1,
gpt-5,o3, and Custom GPTs on December 11, GPT-4o and GPT-4.1 on January 14, 2027 - Self-service fine-tuning is sunsetting: no new jobs after January 6, 2027
- Check the deprecations page monthly: notice periods can drop to two weeks for preview models
Test Your Knowledge
Validate your understanding of the entire course.
1. Which model should you choose to start a demanding new project on November 5, 2026?
Answer: GPT-6 Astra for the most demanding tasks, or GPT-6.1 Sol for near-equivalent performance at a much lower cost ($2 / $10 per million tokens). GPT-6 Luna suits high-volume workloads.
2. Why might a request with `reasoning.effort: "none"` fail on GPT-6.1 Sol?
Answer: Because GPT-6 Astra and GPT-6.1 Sol do not accept none effort. You must switch to low, or choose a model that accepts it, such as GPT-6.1 Terra or GPT-6 Luna.
3. You switch from `gpt-5.4` to `gpt-5.6-terra` without specifying reasoning effort: what happens?
Answer: gpt-5.4 did not reason by default, whereas GPT-5.6 starts at medium. The model will therefore start reasoning, increasing latency and billed tokens. Explicitly set the effort in every request.
4. What happens to Custom GPTs?
Answer: They will be retired on December 11, 2026, or February 11, 2027 with a waiver. The “Migrate to plugin” option converts instructions into skills, files into reference files, and keeps apps; custom actions must be rebuilt, for example using an MCP server.
5. Your application uses Agent Builder and the Evals platform: what is the deadline?
Answer: November 30, 2026 for both. Migrate agents to the Agents SDK or ChatGPT workspace agents, and your evaluations to a tool like Promptfoo.