Back to news
researchAnthropicOpenAI2026-11-16

Anthropic Shows Claude Fable 5.2 Painting With Simulated Brushes via Python Code

Claude Fable 5.2 paints by writing Python code that simulates brushes on a virtual canvas rather than calling an image model. On Anthropic’s internal robot simulation benchmark it scored 64.9 percent against 29.8 percent for GPT-6 Astra.

Anthropic demonstrated Claude Fable 5.2 painting by generating Python code that simulates brushes, paint and canvas instead of calling an image model. One artwork took about 700,000 simulated strokes, and the model scored 64.9 percent on Anthropic’s internal robot simulation benchmark.

ClaudeFable 5.2代码绘图机器人