[FEATURE] Stable vs bleeding-edge release channels for code AND server-side settings

Resolved 💬 1 comment Opened Apr 23, 2026 by giacomov Closed May 28, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

The recent degradation in performance following prompt and server-side settings changes affected many users unexpectedly.

Production-grade applications relying on Claude saw degradation without any obvious explanation and no available fix for weeks.

Proposed Solution

There should be a two-tiers system (modeled loosely against say Ubuntu's "rolling releases" vs stable releases) that encompasses model versions (even micro-versions), code versions (claude code), server-side settings (like defaults for effort and so on):

  • bleeding-edge: the current channel where updates can happen at any time with no guarantee. Users on this channel get all the latest features/models/settings the moment they are released, but also get the related risk
  • stable: after a given period in bleeding-edge, a set of features/model versions/settings is deemed stable and it is available here. Users on this channel get features/models/server-side settings only after they've been battle tested.

IMHO this would give users a way to pick their poison. For critical, production-grade applications, stability is more important than the latest-and-greatest feature.

Alternative Solutions

A user can stay on an older model or avoid upgrading Claude Code or the Agent SDK. However, it is still not protected from server-side changes (model micro-versions, default settings changes, and so on)

Priority

High - Significant impact on productivity

Feature Category

Configuration and settings

Use Case Example

Production-grade software factory: in production I want to use the stable channel so I have a guarantee on the performance of my harness and the stability of its behavior. I can periodically run evals on the bleeding-edge channel to catch early potential future issues and report them to Anthropic, without affecting production

Additional Context

https://www.anthropic.com/engineering/april-23-postmortem

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗