Migrating from gpt-realtime and gpt-4o-realtime to gpt-realtime-2.1
gpt-realtime and gpt-4o-realtime stops answering on Jan 20, 2027 on the OpenAI API. Requests to a retired model fail. The provider’s replacement is gpt-realtime-2.1.
- Shutdown
- Jan 20, 2027
- Time left
- 79 days left
- Replacement
- gpt-realtime-2.1
What shuts down
gpt-realtimeand its dated snapshotsgpt-4o-realtimeand its dated snapshots
Source: developers.openai.com. Check your provider’s deprecations page before you plan around a date.
Find every place you use it
The free scanner flags every reference in your repository and opens one issue with the deadline:
npx swapvera-scan .Or on every push, as a GitHub Action:
- uses: pixelric-spec/swapvera-scanner@v1
with:
fail-on: retiredWhat can change when you swap the ID
Prompts tuned for one model rarely transfer perfectly to the next. Check these before you merge:
- Structured output: JSON that stays valid and keeps the same field types.
- Tool calls: the same tools called with the same arguments, and no calls where there were none.
- Instructions in your system prompt that the new model follows differently, such as length limits or policies it must quote.
- Refusals and safety behavior on your edge cases.
- Parameters the new model rejects, such as a fixed temperature on reasoning models.
- Latency and cost per call at your real input sizes.
What broke for other teams
0of 5 teams
No team has shared results for gpt-realtime and gpt-4o-realtime to gpt-realtime-2.1 yet. Nothing about a migration is published until 5 different teams have shared it, so no single team’s results can be picked out.
Add yours when you run the verdict: swapvera run --share. Only categories are sent, never prompts or outputs. See exactly what is sent.
Check the switch on your own prompts
Swapvera runs your cases on gpt-realtime and gpt-4o-realtime and gpt-realtime-2.1 and comments a verdict on the pull request: safe to migrate, review these cases, or do not migrate.