Migrating off Assistants API — retirement date and replacement
Source last checked: 2026-09-30.
OpenAI is retiring Assistants API on 2026-08-26. The provider’s documented replacement is Responses API and Conversations API. This page is built from OpenAI’s official deprecation notice and is rechecked weekly; the source link is below.
Past scheduled retirement. The published shutdown date has passed; this is a historical notice, not a live availability check.
When is Assistants API being deprecated?
| Model | Provider | Retires | Replacement |
|---|---|---|---|
| Assistants API | OpenAI | 2026-08-26 | Responses API and Conversations API |
Source: OpenAI deprecation notice · announced 2025-08-20
Test before switching
- Check the replacement's supported API endpoints and request parameters.
- Test your prompts, structured output, streaming and tool calls on staging.
- Compare quality, latency, pricing and rate limits on your workload.
- Keep a rollback path while the old model remains available.
Download retirement dates for your models.
What should I migrate Assistants API to?
OpenAI documents Responses API and Conversations API as the successor. This is not a compatibility guarantee: request parameters, API endpoints, pricing, tool calls and output behaviour may differ. Check the linked provider documentation and run your own workload before switching.
Measured latency of the OpenAI API host
| Region | p50 TTFB | p95 | Uptime | Probes |
|---|---|---|---|---|
| Asia (Tokyo) | 215 ms | 296 ms | 100% | 287 |
| Europe (Germany) | 290 ms | 403 ms | 100% | 289 |
| South America (São Paulo) | 262 ms | 487 ms | 99.7% | 288 |
| US (Central) | 134 ms | 224 ms | 100% | 287 |
Edge latency to the provider’s API host over the last 24 hours, measured from 4 regions. This measures the endpoint, not the model: swapping models on the same host is not an inference benchmark or a promise about the replacement model. Compare edge latency →
Other OpenAI models being retired
| Model | Provider | Retires | Replacement |
|---|---|---|---|
| gpt-5.4-cyber | OpenAI | 2026-10-01 | The most capable cyber model available to you. |
| whisper-1 | OpenAI | 2027-02-26 | gpt-live-transcribe or gpt-transcribe |
| gpt-4o-transcribe | OpenAI | 2027-02-26 | gpt-live-transcribe or gpt-transcribe |
| gpt-4o-mini-transcribe | OpenAI | 2027-02-26 | gpt-live-transcribe or gpt-transcribe |
| gpt-4o-transcribe-diarize | OpenAI | 2027-02-26 | gpt-live-transcribe or gpt-transcribe |
| gpt-realtime | OpenAI | 2027-01-20 | gpt-realtime-2.1 |
| gpt-audio | OpenAI | 2027-01-20 | gpt-audio-1.5 |
| gpt-4o-audio | OpenAI | 2027-01-20 | gpt-audio-1.5 |
How is this calendar maintained?
Entries are pulled from provider documentation by an automated weekly job and every one carries a link to its source. Nothing is inferred from blog posts or social media. When a provider changes a date, the change appears here after the next run and is visible in the changelog. The full calendar across all providers is on the deprecations page, and this provider’s own list is at OpenAI deprecations.