Serverless deprecation and migration FAQ

If a serverless model is deprecated, requests may eventually fail or become unavailable on some routes.

Use this checklist to migrate safely.

Quick steps

1. Confirm deprecation status in official docs.

2. Choose a replacement model based on capability, latency, and cost needs.

3. Run a small A/B test before full cutover.

4. Shift traffic gradually (canary -> partial -> full).

5. Keep a fallback path during rollout if possible.

FAQ

"Model is listed, why do I get disabled/unavailable?"

Catalog visibility and route-level availability can differ during transitions.

Runtime/API behavior plus deprecation guidance should be treated as current source of truth.

"Can support re-enable deprecated model for my account?"

In most cases, no immediate per-account re-enable is available through normal support workflows.

"Will replacement model behave exactly the same?"

Not always. Validate quality, latency, and cost on representative workloads before full migration.

"How do I avoid production disruption?"

Use staged rollout and monitor error/latency/cost during migration.

Support intake checklist

- current model ID

- endpoint/route used

- request IDs + UTC timestamps

- exact error payload

- production impact level

Reference

- [Deprecations](https://docs.together.ai/docs/deprecations)