OpenAI removes the Assistants API on August 26, 2026. Its official migration guide points at prompt objects, which OpenAI deprecated on June 3 and switches off 96 days after the deadline that forces the move.
The migration guide states the deadline, "It will shut down on August 26, 2026," and gives a four-row mapping: Assistants become Prompts, Threads become Conversations, Runs become Responses, Run steps become Items. Three of those rows land on objects that are current. One does not.
The prompt object is already deprecated
OpenAI's deprecations page records that on June 3, 2026 it notified developers using reusable prompts "in the dashboard and API" that those objects are being deprecated, with v1/prompts shutting down on November 30, 2026. The prompting guide repeats it: "Prompt creation will be de-emphasized beginning June 3, 2026."
Both statements were live on OpenAI's documentation site on August 20, 2026, at the same time as the migration guide recommending the object.
August 26 to November 30 is 96 days. A team that follows the mapping literally this week adopts an object, then migrates off it inside a quarter.
There is a second reason to skip it
The migration guide also notes a property of the replacement that matters to anyone who provisions infrastructure from code: "Their replacement, prompts, can only be created in the dashboard."
The Assistant it replaces was a persistent API object. So the mapped path converts something a deployment script could create into something a person clicks, and then removes it in November anyway.
The deprecation notice for prompts names the destination directly: move reusable prompt content into your application code. That is reachable in one hop. Instructions and tool schemas live in source control, get passed per request, and state is carried by previous_response_id or the Conversations API, both documented in the Responses migration guide.
One default inverts, quietly
The Responses migration guide lists the mechanical differences you would expect: a typed output array in place of choices, structured outputs moving to text.format, no n for multiple generations. Those announce themselves the first time you run the code.
One does not. In Chat Completions, functions are non-strict by default. In Responses, omitting strict attempts strict mode. Nothing errors at the call site and nothing changes in the shape of your request, so the failure surfaces on whichever real payload first violates the schema you never tightened on purpose. A migration checklist built from the visible differences will not catch it, and a passing test suite is not evidence that the behaviour is right.
What this is and is not
This is a documented comparison of four OpenAI pages read on August 20, 2026, with the shutdown date and the dashboard-only sentence re-confirmed against the live guide at publication. No API call was made and no account was used, so nothing here measures whether a prompt object still functions today, how long a migration takes, or how either path performs. The plainest explanation for the mismatch is also the least dramatic one: the migration guide predates the June deprecation and has not been revised. That is a documentation lag rather than a plan, and it costs the reader the same amount either way.
One absence is worth naming because it affects planning. Neither the migration guide nor the deprecations page documents what happens to existing Assistants and Threads data after the shutdown. There is no stated export tool and no stated retention window. The guide only recommends migrating new user threads onto conversations and older ones as necessary.
If you still have an Assistants integration, the useful takeaway is not urgency, which the calendar already supplies. It is that the documented mapping has one row you should not follow, and the vendor's own later notice tells you where to go instead.