Grok 4.5 Nears GPT-5.6 in Physics Demo Eval

rohanpaul_ai · x · 2026-07-11

A comparative test using atomic chat reveals that Grok 4.5 performed at the level of GPT-5.6 Sol on 3 browser physics demo tasks, and it is free to use.

The test required models to generate self-contained HTML5 canvas code for three scenarios:

The author emphasizes that such tasks expose weaknesses not easily seen in standard coding benchmarks: models must not only generate substantial code but also consistently handle gravity, collisions, velocity, and frame-to-frame behavioral consistency. The post also compares token count, time, and cost across several models, with Grok 4.5 performing remarkably close to GPT-5.6.

Related event: HTML5 Physics Simulation Tests Show GPT-5.6 Pricier but Not Superior(5 posts)→

Original post →

More from Models

Models channel →