# Together AI adds canary rollouts to Dedicated Model Inference

Dedicated Model Inference can now upgrade the model behind a live endpoint without downtime. Traffic shifts in gated steps, and metric gates can pause a rollout if p95 latency or the error rate worsens.

Language: en
Time zone: America/Los_Angeles
Current time (UTC, cached as of): 2026-11-10T06:08:00.000Z
Query date (America/Los_Angeles): 2026-11-09

HTML: https://didcodexreset.com/news/6f141e68ab0844d18e0ac466.html

Content language: en
Localization state: sameLanguage

Source: Together AI · Published 9/22/2026, 17:03:44

Together AI has added canary rollouts to Dedicated Model Inference, so the model behind a live endpoint can be upgraded without downtime. Traffic moves from the current deployment to a new checkpoint in gated steps, defaulting to 5%, 25%, 50%, and 100%.

Health checks run before any traffic shifts. After each step, metric gates compare the new model’s p95 latency and error rate with the previous deployment. If a gate trips, the rollout pauses at the canary share until it is resumed, promoted to 100%, or rolled back.

Canary, blue-green, and rolling strategies are available now through the tg CLI, REST API, and Python SDK. Together AI explains how to start a rollout in [Canary rollouts: upgrade models in production without downtime](https://www.together.ai/blog/canary-rollouts-upgrade-models-in-production-without-downtime).

Tags: Together AI, Dedicated Model Inference, Canary rollouts, Model deployment

[View original post](https://x.com/togethercompute/status/2102546938361262380)