Release: llm-anthropic 0.31

Why it matters

Anthropic released two new models, Fable 5.2 and Mythos 5.2, and an early hands-on test has already surfaced a notable flaw. The author evaluated Fable 5.2's image generation by prompting it for a pelican illustration across five reasoning effort settings. While lower and mid-range settings produced results, with the strongest output at the second-highest level, the maximum setting failed entirely: the model exhausted its entire thinking token budget without delivering any image. This failure mirrors a bug previously observed in Opus 5.5, but it is new to the Fable line, since both Fable 5 and Fable 5.1 handled maximum-effort requests successfully. The regression suggests Anthropic's highest reasoning modes remain unreliable for image tasks even as the models evolve. The article also documents practical access: the models can be run through the llm-anthropic plugin, with installation steps and an example prompt included.