Claude Fable 5.2 gave no image at max reasoning in a test

Anthropic released Claude Fable 5.2 and Mythos 5.2. On the same day, a developer added the two to llm-anthropic, software that lets users send prompts to Anthropic models. The developer did a pelican drawing test at five reasoning levels. The top level, max, gave no image. The model used 128,000 tokens on thinking and then had no tokens left.

How it works

The developer used one test at five levels: low, medium, high, xhigh and max. Each level up makes the model think for more time before it gives output. The result that the developer shows came from xhigh.

The numbers

One of the five runs gave no image. This was the max level. The developer saw the same bug with Opus 5.5. It is new for the Fable series, because Fable 5 and Fable 5.1 gave max level images.

What is not known

This is one test by one person. It does not show how frequently the bug occurs. The source has no word from Anthropic. Anthropic gave no test scores for coding, reasoning or vision at launch.

Insight: The top mode can cost most and give nothing

Anthropic lowered the Fable input price to $8 for each million tokens. The output price is $40. One test with 128,000 tokens on thinking has a price of about $5. This is true if the price for thinking is the output price. The lower price reads as good news, but the top mode can use the most tokens for no result. Test the top mode on your tasks before you use it.