Day: March 26, 2025

In a recent benchmark test, DeepSeek V3-0324 has shown significant improvement, now matching Sonnet 3.7 in terms of code creativity and complexity. The test involved asking various language models (LLMs) to write a raytracer in Python, capable of rendering a visually interesting scene with many colorful light sources. The goal was to evaluate how well the models could generate not only correct code but also produce aesthetically pleasing and varied outputs.