Anthropic releases Fable 5.2 and Mythos 5.2; llm-anthropic 0.31 adds Fable 5.2 support

Anthropic released Fable 5.2 and Mythos 5.2, and version 0.31 of the llm-anthropic plugin adds support for Fable 5.2, according to a post on Simon Willison's blog dated November 9, 2026. Anthropic's announcement is titled "Claude Fable and Mythos 5.2."

Running Fable 5.2 through LLM

Willison's llm-anthropic plugin lets users call Anthropic models from the llm command-line tool. The model ID for the new release is claude-fable-5.2. Setup is:

uv tool install llm
llm install llm-anthropic
llm keys set anthropic
# paste key here
llm -m claude-fable-5.2 "Generate an SVG of a pelican riding a bicycle"

Five reasoning levels tested

Willison ran his pelican-on-a-bicycle SVG prompt at all five reasoning levels available for Fable 5.2: low, medium, high, xhigh and max. He published all outputs in a public gist.

His best result came from xhigh. He describes it as good: the legs are on the correct side of the bicycle, the frame is sound and the pouch is well drawn. He also notes odd small fingers on the handlebars and a pouch that sits slightly out of alignment with the top of the beak.

Max effort failure

The max level did not return an image. According to Willison, the model ran out of tokens after spending 128,000 tokens on reasoning. He reports seeing the same behavior with Opus 5.5 at max effort, which he wrote up in September 2026.

He says this is new for the Fable series. Fable 5 and Fable 5.1 both returned max-effort pelican images without problems in his earlier tests.

Context window, pricing and benchmark figures for Fable 5.2 and Mythos 5.2 are not covered in Willison's post. Details on Mythos 5.2 are in Anthropic's announcement.

What this means

For developers using llm-anthropic, Fable 5.2 is available immediately through the claude-fable-5.2 model ID after upgrading to 0.31.

The max-effort result is a practical caution. In Willison's test, the highest reasoning setting consumed its entire token budget on thinking and produced no output, while xhigh produced the best image. Anyone moving existing Fable 5.1 workloads to Fable 5.2 at max effort should check that responses complete, since the same prompt-level setting behaved differently in the earlier versions. This is a single prompt tested by one developer, not a systematic evaluation.