What changed in your AI tools?

A client update can change the experience.

Tools, permissions, context handling and default settings can change independently of model weights. Before calling a different answer a model downgrade, record your installed client version, selected model and reasoning settings. Check current incidents and compare a repeatable task using the quality investigation guide.

What is monitored automatically?

We check the latest stable GitHub release from OpenAI’s Codex repository and Anthropic’s Claude Code repository about every six hours. If the GitHub API is unavailable, we verify the official stable-release page and its publication time. Failed checks retry on the next collector cycle. Preview, alpha and draft releases are excluded. This is a latest-release watch: it can miss versions published and replaced between checks. Previously observed releases remain in the feed.

What needs editorial review?

“Change explained” and historical postmortems are selected, source-linked notes reviewed when published here. This first version does not automatically monitor every model retirement, quota policy, product announcement or X post. A new version is not proof that quality improved or declined.

New stable-release browser alerts start after our initial collection; historical entries are not sent as new announcements. Subscribe to Changes RSS ↗