As regular readers of this blog will know, I have been using Claude Haiku 5.5 and GPT-6-Luna on OpenRouter to generate Rust code. Most of the time, code generation costs less than $0.01 (often as low as $0.005) with both of these models, so they are economical assuming you use them only a few times a day.
Both models also seem to respond with the generated content in 10-30 seconds, so they are very similar overall. I have noticed Claude Haiku has a little different style of Rust code it generates—it tends to prefer separate functions, while GPT-6-Luna puts more things in a single function.
As far as functionality goes, the code usually works on my first attempt in both models. Sometimes Haiku or Luna will write code that does not fit my needs correctly. Usually a more advanced model like Opus or Sol will generate correct code in these cases.
So the comparison results: both of these "cheap" models occasionally make mistakes, but they are often good enough, and if one fails, try the other. As far as when to use Haiku and Luna—I would recommend using them first on all code generation, and only if the result is incorrect, using a more expensive model.