Descriptions:
Fahd Mirza puts two flagship closed models head-to-head — Claude Fable 5 from Anthropic and GPT-5.6 Sol from OpenAI — using a single deliberately brutal prompt rather than a cherry-picked battery of tests. The challenge: generate a procedural animation of a rotating döner kebab with realistic fat-droplet physics, color-temperature mapping from white-hot to red as droplets fall, camera movement, and a seamless 8-second loop — all in pure vanilla JavaScript, one file, zero libraries.
Before revealing results, Mirza lays out the key spec differences between the two models. Claude Fable 5 is priced at $10 input / $50 output per million tokens and carries a 1 million token context window, with SWE-bench Pro scores of 80.3%. GPT-5.6 Sol comes in cheaper at $5 input / $30 output, offers a 400,000 token context window, and scores 64.6% on SWE-bench Pro — but edges ahead on coding agent index, browser comparison benchmarks, and terminal bench, suggesting stronger agentic efficiency for tool-heavy tasks.
The live outputs track closely with these benchmark profiles. Sol produces more geometrically convincing kebab geometry and surface shading, while Fable 5 demonstrates stronger physics reasoning in parts of the simulation but falls short on instruction-following for specific visual details like the tray mandala. The video is a useful reference for developers choosing between these models for complex, constraint-heavy code generation work.
📺 Source: Fahd Mirza · Published July 12, 2026
🏷️ Format: Comparison
