Comparing Models for Parametric Furniture Modeling

TLDR: I think Fable 5 builds best model with relatively low cost. Not a benchmark, but I think the result is interesting. I used each of these LLMs to build furniture model with identical prompt. And compare the build time, token usage, cost and model quality. GPT 5.5 is fast, token efficient, but it wasn't following the prompt, the dovetails are not at the corners and joinery it invented is physically impossible to be assembled. (I highlighted the broken joint in each of the images). Fable 5 is still fast,