Anthropic lists which Claude models carry the text watermark

Published 20 September 2026

Anthropic's help article on how Claude marks generated content now carries a table of which model does what. Three models carry the text watermark: Claude Fable 5.1, Mythos 5.1 and Opus 5. Anthropic says marking is added at launch for models released on or after 2 August 2026, and that it is working to extend marking to earlier models during a transition period[1]. Until this table appeared, the article said only that new models were covered, without naming them.

What changed

The help article was updated on 16 September 2026 with a per-model table. The models named as carrying the text watermark are the three launched after the 2 August cutoff. Marking applies wherever Claude is offered, worldwide, across the API, the Claude platform, Claude Code, Claude Cowork and Claude Tag[1].

What it means

If you use an older model through the API today, its text is not listed as marked. That is a statement about the table as it stands, not a promise: Anthropic describes the older models as covered during a transition period, so the list is expected to grow. If your product depends on knowing whether a given model's output is marked, read the table on the day you need it rather than relying on this page. What this means for developers.

What is still not known

Anthropic has not published detection accuracy figures, and the detector remains in private preview for organisations obliged to check under EU law[10][1]. A listed model's text is marked; whether anyone you deal with can check it is a separate question. The full picture of what Anthropic shipped.

Last verified against primary sources: 2026-09-18