The Vibe Father
GitHub Models is fully retired as of July 30, 2026. GitHub says the playground, model catalog, inference API, and bring-your-own-key endpoints are unavailable to every customer, including accounts with active usage. The shutdown completes the retirement announced on July 1.
What disappeared
- The browser playground for trying models.
- The hosted model catalog.
- GitHub Models inference API endpoints.
- Bring-your-own-key access through the product.
GitHub directs general model-access workloads to Microsoft Foundry and points GitHub-native AI workflows toward Copilot. Those are product suggestions, not automatic migrations.
Migration checklist
| Layer | Check before cutover |
|---|---|
| Endpoint | Base URL, region, model IDs, API version, and streaming format |
| Authentication | Identity method, secret rotation, and least-privilege scope |
| Request schema | Messages or responses format, tools, JSON schema, and token limits |
| Behavior | Representative evals, refusals, output shape, and tool selection |
| Operations | Rate limits, retries, logging, budget alerts, and incident owner |
| Data | Retention, training defaults, geography, and compliance terms |
Do not hide the provider behind a weak abstraction
Portability does not require pretending all models and APIs are identical. Keep a small adapter around transport and model selection, then expose capabilities explicitly. If one provider supports strict structured output and another does not, the application should know.
Save prompts, tool definitions, test cases, and accepted outputs outside a vendor playground. A hosted experiment is useful, but it should not be the only copy of the work required to reproduce a production behavior.
What to test after migration
Run a smoke suite for authentication, streaming, tool calls, structured output, timeouts, and rate-limit handling. Then run task-level evaluations. A 200 response only proves that the new endpoint answered.
Bottom line
The retirement is complete. Any remaining GitHub Models dependency is now an incident, not a future planning item. Migrate the transport, verify application behavior, and use the shutdown to make the next provider change less disruptive.
