DeepCodeBench: Real-World Codebase Understanding by Q&A Benchmarking(qodo.ai)
qodo.ai
DeepCodeBench: Real-World Codebase Understanding by Q&A Benchmarking
https://www.qodo.ai/blog/deepcodebench-real-world-codebase-understanding-by-qa-benchmarking/
7 comments
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
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?
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?
I particularly like their usage of LLM-as-a-judge. They don't go "hey chatgpt, sort these from best to worst based on vibes", rather they extract a set of ground truths and check how the answer compares, a task that SOTA LLM can do kind of reliably. It's a very smart way to circumvent the problems introduced by pure LLM-as-a-judge methods.
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.