Claude Fable 5.1 made me a really nice animated pelican

Anthropic released Claude Fable 5.1, boasting major improvements in coding, knowledge work, and long-term problem-solving. Its standout achievement is a 52.6 percent score on the Terminal-Bench-Science 0.1 benchmark, outperforming previous models by a wide margin.

The model features five reasoning effort levels ranging from low to max, with no option to turn reasoning completely off. Testing the model with a prompt to generate an SVG of a pelican riding a bicycle reveals how output changes across these effort levels. At low and medium settings, the model skips reasoning entirely and produces basic results within seconds at a fraction of a cent.

Higher settings dramatically increase token usage, processing time, and cost while generating much more detailed outputs. The extra high setting takes nearly eight minutes and costs $1.83, producing a lengthy reasoning trace and an intricate design. The max setting takes nearly fourteen minutes and costs $3.30, resulting in the best pelican graphic yet, complete with a tasteful background, precise anatomical details like feet on pedals, a blue hat, and a basket with a fish.

Following a user suggestion on Hacker News to animate the result, the author fed the max pelican back into the model with a prompt to animate it. Using the high reasoning level at a cost of $1.37, the model successfully generated an animated SVG of the pelican riding the bicycle.