Skip to content
Hugin
Railway tracks in fog where one line divides into two that curve apart, cold blue dusk light along the steel rails.

Journal

The retirement that never breaks is the one to watch. The old name keeps answering — and something else is doing the answering.

3 min read

Original editorial artwork generated for Hugin.

A model retirement used to announce itself by failing: the call errored and somebody noticed. This month's schedules show a quieter pattern. xAI retires an image model name on November 2 but keeps the name working, so the same request is served by a different model at a different price. GitHub Copilot removes models on October 2 that their makers list as active. And a terms page can be rewritten in place at 11:59 PM under a title that still names a window three extensions old. None of these break anything. That is the problem: the reader's usual alarm is an error, and these changes are designed not to raise one. The defence is to ask a different question — not 'does it still work?' but 'what is answering when I call this name?'

methoddeprecationsaixaigithub-copilotdated-terms

For as long as software has had versions, retirement had a sound. You called the old thing and it failed. An error in a log, a broken build, a support ticket — the failure was the notice, and it arrived on its own whether or not anyone read the deprecation page.

That alarm is being engineered out, on purpose and often for good reasons. And it is worth being clear about what goes with it.

Three changes that break nothing

A name that keeps answering. xAI retires its grok-imagine-image-quality model name on November 2. Its migration guide is careful to reassure: "The slug continues to resolve, so you do not need to change your code to avoid breakage." What changes is behind the name — requests are served by grok-imagine-image-2.0 at low quality, which "is priced differently than the model it replaces." In this case the price moves in the caller's favour. The point is not the price. The point is that the model producing your output changes on a date, and nothing in your system will tell you.

A model that leaves the tool, not the maker. GitHub Copilot removes Claude Opus 4.7 and three other models on October 2. Anthropic's own deprecations page lists Opus 4.7 as Active. A developer who checks the maker's schedule finds nothing wrong; the model simply stops appearing in the menu they use it from.

A page that changes under its own title. The Claude Code weekly-limits terms were rewritten at 11:59 PM Pacific on the night the window ended, replacing a promise to "return to their standard levels" with a permanent increase. The change was good news. It also happened in place, under a title that still names May–August, on a page whose earlier versions said something else.

Why the old alarm does not catch these

Every one of these was announced — honestly, specifically, with dates. None of them was hidden. But each is built so that the ordinary signal of change, a failure, does not occur. A monitoring habit tuned to errors will pass all three without a flicker, because from the outside the thing still works.

That is a real improvement for the people who would otherwise have been broken. It is a real risk for anyone whose correctness depends on which thing answers: a researcher whose results depend on a specific model, a business whose costs are estimated against a specific rate, a team whose agreement names a specific tool.

The question to ask instead

The breaking-change era trained one question: does it still work? The non-breaking era needs a second one: what is answering when I call this name?

In practice that means three habits:

  • Pin the thing, not the name. Where a provider offers an explicit version or model id, call that. xAI's own recommendation is exactly this: "We recommend switching to grok-imagine-image-2.0 explicitly before November 2."
  • Read the schedule of the tool you use, not only the lab behind it. The maker's page answers a question about the maker.
  • Treat the operative page as the record, and its date as part of the record. A page that is edited in place is a different document after the edit, whatever its title says.

None of these is exotic. They are what a careful reader of a contract already does: read the version that is in force, not the one you remember signing.

Source links