OpenAI gave developers exactly one year of notice, posting the deprecation on 26 August 2025 and setting the shutdown for 26 August 2026 — one day from this article’s publication. There is no automated migration tool for the two objects most integrations depend on — Threads and Assistants — so teams that have not already rebuilt on the Responses API and the Conversations API are migrating by hand, against a deadline that is now one day away. OpenAI’s own migration guide covers three separate conversions: Assistants to Prompts, Threads to Conversations, and Runs to Responses. Vector store objects, which are not unique to the Assistants API, continue to be usable with the Responses API’s file-search tool.

Why it matters

Any production application built on the Assistants API beta is about to call a shut-down endpoint. Unlike a typical deprecation with a soft-fail window, OpenAI has not published a grace period or partial-availability plan for 26 August 2026 — its own documentation describes the date as when the API “will shut down,” full stop. For a team that has not migrated, the practical effect is broken checkout flows, broken support bots, or broken internal tools the moment tomorrow’s traffic hits the old endpoints. For everyone else, this is a live case study in how much of a “beta” label OpenAI is willing to enforce, and how much lead time is actually usable when the fix requires re-architecting how conversation state is stored.

What OpenAI has confirmed, and where

OpenAI’s deprecations page states plainly: “On August 26th, 2025, we notified developers using the Assistants API of its deprecation and removal from the API one year later, on August 26, 2026.” The same page lists the shutdown in its deprecation table as 2026-08-26, with “Responses API and Conversations API” as the recommended replacement. The Assistants file-search documentation carries an identical notice at the top of the page: “After achieving feature parity in the Responses API, we’ve deprecated the Assistants API. It will shut down on August 26, 2026.”

The original decision predates the one-year countdown. When OpenAI launched the Responses API in March 2025, it said it planned to bring all Assistants API functionality into the newer API and would set a 2026 sunset date once that parity was reached. The 26 August 2025 notice was OpenAI confirming that parity had been reached and starting the clock.

The announcement itself was posted to OpenAI’s own developer community forum by a user posting as “edwinarbus,” identified elsewhere on the same forum as OpenAI’s Developer Community Lead — an OpenAI staff account, not a third party. That post states the API “will sunset one year from now, August 26, 2026,” and frames Assistants as “our early take on how agents could be built,” predating the current generation of reasoning models, with its useful parts folded into Responses.

What breaks, and what does not

Three Assistants API objects go away and have no automatic replacement: Assistants, the reusable named configuration of a model, instructions and tools, which must be manually recreated as “prompts” in the OpenAI dashboard with no documented bulk-conversion path; Threads, the server-stored conversation history, for which OpenAI’s migration guide is explicit — “We will not provide an automated tool for migrating Threads to Conversations. Instead, we recommend migrating new user threads onto conversations and migrating older ones as necessary”; and Runs, the execution of an assistant against a thread, which maps conceptually to a Responses API call but requires the calling application to now handle orchestration that used to run on OpenAI’s servers.

Vector stores sit apart from this. They are a separate API resource, referenced by ID from the file-search tool — a pattern that predates the Assistants API’s beta status and appears unchanged in OpenAI’s current Responses API file-search documentation. Neither the deprecations page nor the migration guide singles out vector stores or uploaded files as being deleted alongside Assistants, Threads and Runs. That is a reasonable inference from OpenAI’s own documentation rather than an explicit line saying vector stores will survive this, and readers should treat it accordingly: OpenAI has not published a sentence to that specific effect.

What OpenAI has not published, in any of the pages checked directly for this article, is a description of the exact error behaviour after the cutover: whether calls to the old endpoints return a specific error code, whether there is any tapering period, or whether the shutdown is an instantaneous, all-at-once cutoff. Absent a published grace period, developers should plan for a hard stop rather than assume a soft one.

The pattern fits a broader push toward the Responses API

This shutdown is not an isolated housekeeping move. OpenAI has been steering developer traffic toward the Responses API for over a year — see 18BYTE’s coverage of OpenAI’s price cut on its top model. The Assistants API sunset is the cleanup half of that same strategy: fewer parallel APIs to support, with Responses and Conversations absorbing the built-in tools — code interpreter, file search, deep research, MCP connections and computer use — that Assistants offered piecemeal.

What this means for buyers

If your team, or a vendor you depend on, has a production integration built on the Assistants API, the question to ask today is not “when is the deadline” — it is tomorrow, 26 August 2026 — but “what will start returning errors, and for whom, the moment it arrives.” Ask vendors directly whether their integration used Assistants, Threads and Runs, and if so, request evidence of a completed migration rather than a promise of an imminent one. If vector stores and uploaded files matter to your workflow, separately confirm those specific IDs still resolve correctly through the Responses API’s file-search tool in your vendor’s account, since OpenAI has not published an explicit guarantee on that point.

What would change our reading

If OpenAI publishes a specific error code or a phased rollback plan for the cutover, that would soften the “hard stop, no grace period” reading above. If OpenAI issues an explicit statement confirming vector stores and files created via the Assistants API are guaranteed to remain accessible through the Responses API, that would upgrade the current inference to a confirmed fact.

Sources