Summary
Can GPT 6.2 Astra or Gemini 4 Carbon dethrone Claude Fable 5.2 as the best model available? Peter Yang runs all three through five hands-on tests to find out, sharing his prompts so viewers can repeat the experiments.
The tests include a 20-second rainy-night marketing video of Shibuya Crossing rendered entirely in code, a playable Godzilla-style rampage game built with Three.js and Blender models, and a redesign of a personal website. Generation times vary widely: Astra finished the video in about 11 minutes, while Fable took nearly an hour and Gemini 4 Carbon 73 minutes. Fable wins the video and game tests, with Astra placing ahead of Gemini, while website design is close between Fable and Astra, with Gemini clearly behind.
The final test focuses on writing, which Yang calls the final boss of AI because of the difficulty of avoiding slop. He asks each model for a 300-word personal essay from a software engineer whose craft has been taken over by AI, and compares the results blind. He also tests how well each model holds its position when challenged. The video is a practical guide to choosing between today’s leading models for coding, design and writing, with surprises along the way, including where Gemini comes out on top.
📺 Source: Peter Yang · Published November 18, 2026
🏷️ Format: Comparison







