Two OpenAI shutdowns land on 26 August 2026: o3 leaves ChatGPT and the Assistants API is gone for good
On 26 August 2026, OpenAI removes o3 from ChatGPT's model picker and permanently shuts down the Assistants API. Here's what breaks and what to do.
26 August 2026 is a date worth circling if you use ChatGPT or build on the OpenAI API. Two separate shutdowns land on the same day: o3 disappears from ChatGPT’s model picker, and the Assistants API is permanently switched off. They affect completely different groups of people, so this post covers each one in turn.
For ChatGPT users: o3 is leaving the model picker
If you have been selecting o3 manually in ChatGPT as your go-to reasoning model, that option goes away on 26 August 2026. OpenAI announced a 90-day sunset period, and the deadline is now here.
A few things worth clarifying so there is no confusion:
o3-pro is not going away. If you are on a Pro, Team, Enterprise, or Edu subscription, o3-pro remains available. It is the consumer-grade o3 that is being retired, not the entire o3 line.
The API is unaffected by this change. OpenAI was explicit: the model picker retirement applies to ChatGPT only. If you call o3 through the API in your own applications, you have a separate deadline to work with. The older snapshot o3-2025-04-16 is set for removal on 11 December 2026, per a deprecation notice issued on 11 June 2026.
What to move to. For paid subscribers who relied on o3 for heavy reasoning work, the practical options are o3-pro if your plan includes it, or the GPT-5 family for tasks where you want a capable model without the reasoning overhead. OpenAI’s model selection in ChatGPT has shifted toward these newer options, and the quality gap between o3 and what is available today is narrower than it was at o3’s launch.
This consolidation pattern is consistent. o4-mini was retired from ChatGPT on 13 February 2026 following a notice on 29 January. OpenAI rotates models out of the consumer picker as newer ones become the clear recommendation.
For API developers: the Assistants API is shutting down permanently
This is the bigger disruption, and “deprecated” does not do it justice. On 26 August 2026, all Assistants API endpoints are removed. Requests to /v1/assistants, /v1/threads, /v1/runs, and related endpoints will fail with errors. OpenAI has stated there is no extension option.
The deprecation notice went out on 26 August 2025, giving developers a full year. That year is now up.
Why OpenAI is removing it
The Assistants API was always in beta and never graduated to general availability. The underlying design required three separate API objects (Assistants, Threads, Runs) to accomplish what is fundamentally a “send input, get output” operation. The polling loop for Run completion was awkward, submitting tool outputs required additional API calls, and Threads created context window management bottlenecks at scale. When OpenAI released the Responses API in March 2025, it made the Assistants model look even more convoluted by comparison.
What replaces it
The Responses API is the migration target. It is stateless by default, with optional server-side state available through the separate Conversations API. The conceptual mapping is straightforward:
- Assistants become Prompts, created and versioned in the dashboard rather than managed through API objects
- Threads become Conversations, with opt-in server-side persistence rather than automatic state management
- Runs become Responses, with explicit tool loops rather than the polling model
Beyond simplifying the architecture, the Responses API unlocks features the Assistants API never will: Computer Use, Model Context Protocol (MCP) server connections, background mode for long-running tasks, and encrypted reasoning items for regulated industries. OpenAI has also noted 40 to 80 per cent better cache utilisation, which feeds through to cost.
What you need to do before 26 August 2026
Export your data now. Vector stores and files will migrate to the Responses API’s File Search tool, but assistant definitions and threads do not carry over automatically. After the deadline, that data is gone. Do not wait.
Rebuild your assistant logic as Responses API calls. OpenAI will not provide an automated migration tool for Threads to Conversations. The recommendation is to migrate new user threads onto Conversations going forward, and migrate older threads as needed based on their business value.
Check your third-party integrations. Zapier has deprecated all ChatGPT steps that rely on the Assistants API, including “Conversation With Assistant” and “Create Assistant”. Zaps using these steps will stop working on 26 August 2026 and need to be rebuilt using Responses API actions.
Azure users, check your status. Microsoft’s Azure AI Foundry documentation carries a deprecation notice stating the Azure Assistants API will be retired on 26 August 2026, with migration to the Microsoft Foundry Agents service. Earlier guidance from October 2025 suggested Azure was unaffected; that guidance has since been superseded. If your team has Azure-hosted assistants, verify your migration path against the current Microsoft documentation.
The broader picture
26 August 2026 is one point in a larger wave of OpenAI removals across 2026. ChatGPT Atlas went on 9 August, Imagen 4 on 17 August, and DALL·E GPT follows on 30 August. The larger sweep comes on 23 October 2026, when GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, o1, o1-pro, o3-mini, and o4-mini are all removed from the API in one batch. If your codebase has not been audited for deprecated model calls recently, that October date should be on your roadmap now.
The short version
If you use ChatGPT and selected o3 manually: from 26 August 2026, it will no longer appear in the picker. Move to o3-pro if your plan includes it, or to GPT-5 family models.
If you build on the Assistants API: the endpoints are gone on 26 August 2026 with no extension. Export your data, migrate to the Responses API, and rebuild your thread and assistant logic before that date. The official migration guide is the right starting point.