Fable 5.2 hits llm-anthropic; max effort burns 128,000 tokens

Anthropic's release of Fable 5.2 and Mythos 5.2 arrived with same-day support in llm-anthropic, the plugin that lets developers run prompts through Anthropic models from the command line. Willison's first test was a pelican-on-a-bicycle SVG prompt, run at all five reasoning levels the model exposes—low, medium, high, xhigh, and max. The xhigh setting produced the best image he got.

The max setting behaved differently: it spent 128,000 tokens thinking and then returned no image at all. Willison notes that this is the same bug he saw with Opus 5.5, but it is new for the Fable series, since both Fable 5 and Fable 5.1 returned max effort pelicans just fine. That contrast suggests the failure may be tied to how the newest model allocates its reasoning budget at the top setting, though the post does not say what causes it.

For anyone using llm-anthropic, the practical takeaway is that support for a newly released model can come with rough edges at the highest reasoning level. Willison's test covers a single prompt, so it is not yet clear how broadly the max-effort failure would appear across other tasks.