Used Claude Sonnet 5.5 for Code Gen

A developer reported that Claude Sonnet 5.5 generated two Rust functions for about $0.06 after DeepSeek V4.1 Flash produced hard-to-read code and a slight performance regression. The developer said Sonnet fixed the regression and yielded more readable code, calling it clearly superior to DeepSeek V4.1 Flash at only 2-3 times the cost.

I found two more functions in my Rust program that could be replaced with specifications and generated with AI . I decided to test some AI chat bots to see if I could improvement my implementations. First I tried DeepSeek V4.1 Flash, but it gave me an implementation that was hard to read, almost like spaghetti code . It also seemed to cause a tiny performance regression. I decided to try a different model. I tried the specification prompts in Claude Sonnet 5.5 . This fixed the performance regression, and yielded better, more readable code. So I ended up spending about $0.06 on Sonnet today to generate 2 Rust functions. In this case, Sonnet was clearly superior to DeepSeek V4.1 Flash, and only 2-3 times more expensive. It seems that mass copyright infringement has really helped Anthropic improve their models.