Claude Fable 5.1 made me a nice animated pelican (simonwillison.net)

Anthropic has announced the release of Claude Fable 5.1, which sets a new benchmark in coding and longer problem-solving tasks, showcasing a remarkable score of 52.6% on the new Terminal-Bench-Science 0.1 test. This represents a significant leap from previous models, including Fable 5 and GPT-5.6 Sol, highlighting advancements in scientific research capabilities for AI systems. Notably, Fable 5.1 introduces five reasoning levels—low to max—allowing for varying degrees of detail in outputs. In a compelling demonstration, the model was tasked with generating an SVG of a pelican riding a bicycle, with varying results based on the reasoning level selected. Outputs ranged dramatically in detail and complexity, from simple sketches at low levels to intricate designs at maximum effort, featuring thoughtful considerations like the positioning of a pelican’s legs and adding cosmetic details. The model’s approach emphasizes the ongoing evolution of AI in creative tasks, suggesting that while Fable 5.1 excels in generating structured content, there is still room for improvement compared to competing models like Gemini 3.7 Flash. This release underlines the importance of reasoning capabilities in AI, potentially reshaping workflows in creative industries and research applications.