If you do a bit of digging into most of the popular benchmarks that all the big labs report on, you'll see pretty quickly that they have almost zero correlation with any real world tasks.

The approach that they're taking here of working backwards from a OS repo pull request and reverse engineering a question is unusually well thought out for a benchmark.

I haven't dug into more of the dataset questions yet, but the example they give in the blog post for the question generated for Hugging Face Transformer's repo gives me hope that this could actually be a solid benchmark:

> How do the fast image and video processor base classes prevent shared mutable state when instantiating multiple instances?

Seems like an interesting benchmark, but my takeaway from the results is that Codex is almost as good enough as their custom solution (no mention of the underlying model) and only requires a $20 ChatGPT subscription to start using it (of course with limits), without having to shell out $$$ for an enterprise Qodo plan to use Qodo Aware - https://www.qodo.ai/products/qodo-aware/. The "free" plan in Qodo Aware only lets users work with 100 hand-picked open-source repositories.

It also would be nice if the article clearly mentioned what specific model settings were used for Claude Code and Codex. Both of those allow changing the reasoning level, so if the benchmark was done using the default settings, it seems a little unfair - they have a result of their own agent at high reasoning as a separate entry.

This is in relation to their newly-announced "context agent": https://www.qodo.ai/blog/introducing-qodo-aware-deep-codebas...

Agentic search is good enough for code search and code understanding, indexing/fancy techniques will only slight outperform for a lot more effort