Azure OpenAI's Assistants API Retired August 26, 2026 —
What Broke, and What Didn't Migrate
Microsoft's documentation says it in one sentence, tucked into a note box: "The Assistants API is deprecated and will be retired on August 26, 2026." That deadline — August 26, 2026 — has now passed: it landed in the last week of August, when half the teams who needed to act were on vacation. The replacement is the generally available Microsoft Foundry Agents service, and Microsoft publishes a migration guide and an automated migration tool.
If your application creates assistants, appends messages to threads, or executes runs against Azure OpenAI, that deadline has arrived. And unlike the end-of-life dates we usually track — where unsupported software keeps running, quietly accumulating risk — a retired cloud API is a hard stop. There is no "running unpatched" after August 26, 2026. There is only running, or not — and if your integration still answers today, treat that as borrowed time rather than a reprieve.
What actually retires
The Assistants API was Azure OpenAI's stateful application layer: Assistants (model + instructions + tools), Threads (server-side persistent conversations), Runs (executing an assistant against a thread), plus hosted tools like code interpreter and file search. Its pitch was that Microsoft managed your conversation state so you didn't have to — persistent threads with automatic context-window truncation were the headline feature.
All of it is superseded by Foundry Agents, which restructures the same ideas: threads become conversations (storing messages, tool calls, and outputs as items), and runs become responses — a simpler request/response pattern on the OpenAI client. The concepts map; the SDK calls do not. Code written against client.agents.threads and friends has to change.
The other half: OpenAI is shutting its own Assistants API down the same day
This is the part most coverage misses, and it reaches far more developers than the Azure edition. OpenAI's own Assistants API — the one on platform.openai.com, not Azure — retires on the same date, August 26, 2026. That is not a coincidence: OpenAI announced it on August 26, 2025 with exactly one year of notice, and the Azure retirement is aligned to it.
OpenAI's replacement is a pair of APIs rather than one: the Responses API (the request/response call that superseded Assistants' runs) and the Conversations API (which holds the persistent state Threads used to). In OpenAI's own framing, when the Responses API shipped in March 2025 they committed to bringing every Assistants feature across to it, with the sunset landing in 2026. They publish an Assistants-to-Conversations migration guide alongside the deprecation notice.
The practical consequence for anyone running both: you cannot solve this once. The Azure path moves you onto Foundry Agents; the OpenAI path moves you onto Responses plus Conversations. Different SDK surfaces, different object models, each needing its own testing pass.
Worth noting while you are in the deprecation page: OpenAI has a large retirement wave right behind this one. October 23, 2026 removes o1, o3-mini, o4-mini, gpt-4.1-nano and gpt-3.5-turbo-0125; December 11, 2026 removes o3 and the original GPT-5 snapshots. Every dated OpenAI model retirement is tracked on our OpenAI API lifecycle page.
The detail that catches teams out: your data doesn't come along
Microsoft's migration tool rewrites code — agent definitions, thread creation, message creation, run creation. But the migration guide is explicit about its limit: "It doesn't migrate state data like past runs, threads, or messages." And the guide's own caveat table confirms the consequence: old thread data isn't available in the new service; teams should "start new conversations after migration."
The migration checklist
1. Find your exposure. Search your codebases for Assistants API calls — assistant creation, thread operations, run polling. This includes the quiet internal tools nobody has deployed in months; they break just as hard as production.
2. Run the migration tool, then treat its output as a draft. The automated tool handles the mechanical SDK translation. Tool configurations (code interpreter, file search, function calling) deserve hand-verification — the hosted-tool semantics are where subtle behavior differences live.
3. Export any thread history you can still reach. Pull the messages you must retain via the existing API for as long as it answers. Compliance-sensitive teams: conversation records your retention policy covers are state data, and state data does not migrate.
4. Re-test before you cut over, not after. The failure mode this retirement kept producing was a late cutover that surfaced a tool-mapping issue with no runway left to fix it — worth remembering for the next API on this cadence.
The pattern behind the deadline
This retirement is not an outlier; it is the cadence of the AI platform era. GPT-3.5 Turbo retired on a schedule that would have seemed absurd for a database. AWS deprecates Lambda runtimes on clocks measured in months. The Assistants API went from launch to retirement inside roughly two and a half years — and its successor will iterate on a similar clock. We track AI model and platform deprecations across providers on our AI models lifecycle page precisely because this category now turns over faster than any other software we monitor. The operational lesson for teams building on AI platforms: treat API lifecycles as dependencies with expiry dates, the same way you treat runtimes and operating systems — because the vendors already do.
Frequently Asked Questions
When does the Azure OpenAI Assistants API retire?
August 26, 2026 — stated verbatim in Microsoft's documentation. The recommended replacement is the generally available Microsoft Foundry Agents service, with a published migration guide and automated migration tool.
Is OpenAI retiring its own Assistants API too?
Yes — on the same day. OpenAI's Assistants API on platform.openai.com also shuts down on August 26, 2026, announced a year in advance on August 26, 2025. The replacements are the Responses API (for what runs did) and the Conversations API (for what Threads did), with an official Assistants-to-Conversations migration guide. Anyone building on both Azure and OpenAI has two migrations and one deadline.
Does Microsoft's migration tool move my Assistants data?
No. It migrates code constructs — agent definitions, thread creation, run creation — but Microsoft states it "doesn't migrate state data like past runs, threads, or messages." Conversation history your application depends on must be exported before the retirement date.
What replaces threads and runs?
Threads become conversations (storing messages, tool calls, and tool outputs as items); runs become responses, a simpler request/response pattern usable alone or with conversations for context.
What happens to applications still calling the API after August 26?
A retired cloud API stops serving — expect failures on Assistants, Threads, and Runs calls once retirement takes effect. Unlike on-premises software past end of life, there is no degraded-but-running state.