Karpathy Tests Claude Opus on 3D Movie Generation

Episode 1 · Karpathy's LOTR 3D test pushes long-horizon model evaluation (2026-08-02, 7 posts)

Multiple posts say Andrej Karpathy ran a "Lord of the Rings"-style stress test for frontier models: he provided the opening passage of The Lord of the Rings, a budget of about 1 million tokens (around $10), and asked the model to render the text with Three.js. According to these writeups, the model worked autonomously for roughly 2 hours, produced about 5,500 lines of code, and generated a rough but interactive 3D/game-like scene. The result drew attention because it suggests strong spatial and long-horizon coding ability, while also exposing a key weakness: the model could not reliably verify its own visual output.

Confirmed

• TAbrodi, 机器之心, AI寒武纪 and others describe the same core setup: the model received the opening text of The Lord of the Rings plus a 1 million-token budget, and was tasked with building a Three.js rendering.
• These posts consistently report roughly 2 hours of autonomous work and about 5,500 lines of Three.js code.
• 机器之心 and 新智元 say the final output was an interactive but rough 3D scene or game-like experience.

Unconfirmed

• The exact model name is inconsistent across posts. TAbrodi and AI寒武纪 describe it as Claude 3 Opus, while 量子位 and 新智元 refer to Opus 5. Based on the provided materials alone, the precise version cannot be confirmed.

Why it matters

• andrewchen's repost says Karpathy used the demo to argue that model evaluation is moving beyond simple static tests such as generating a pelican-on-a-bike SVG toward longer, more autonomous tasks.
• AI寒武纪 and 新智元 emphasize the main bottleneck revealed here: even when a model can generate substantial code and plausible 3D structure, it still lacks a robust way to visually inspect, validate, and correct its own output.

• Karpathy Tests Claude Opus: Writes 5,500 Lines of Code in 2 Hours to 3D Render LOTR — andrewchen · 2026-08-02
• Karpathy Tests Opus: Writes 5,500 Lines of Three.js to Render LOTR in 2 Hours — TAbrodi · 2026-08-02
• Karpathy Tests Opus 3D World Generation: Spatial Skills Impress, But Lacks Visual Feedback — AI寒武纪 · 2026-08-02
• Karpathy Tests Claude 3 Opus: Writes 5,500 Lines of 3D Code in 2 Hours for $10 — TAbrodi · 2026-08-02
• Karpathy Uses Opus 5 to Generate 3D LOTR, Exposes LLM Self-Review Limits — 新智元 · 2026-08-03
• Karpathy's New 'LOTR' Benchmark: Opus 5 Burns 1M Tokens to Build 3D World — 量子位 · 2026-08-03
• Karpathy Tests Opus: Generates 3D World with 5,500 Lines of Code for $10 — 机器之心 · 2026-08-03

Episode 2 · Karpathy Tests Claude Opus to Generate 3D Movie (2026-08-04, 2 posts)

Andrej Karpathy recently challenged Claude Opus with a $10 budget to generate a 3D movie scene based on The Lord of the Rings. In just two hours, the AI model successfully generated 5,500 lines of Three.js code, demonstrating its powerful capabilities in complex code generation and 3D rendering.

• Karpathy Tests Claude Opus: Writes 5,500 Lines of Code in 2 Hours to Generate 3D Film — iamrobotbear · 2026-08-04
• Karpathy Tests Claude Opus: Writes 5500 Lines of Code to Render 3D Scenes — jon_barron · 2026-08-05