Start front loading the models with 5k, 10k, 50k, 100k tokens of messy quasi related context, and then run the benchmarks.

These models are ridiculously powerful with a blank slate. It's when they get loaded down with all the necessary (and inevitably unnecessary) context to complete the task that they really start to crumble and fold.

We can definitely make harder evals, the problem is a good eval set is indistinguishable from good training data / market edge, so no one is incentivized to share their best eval sets publicly.

This is the least true thing ever. All LLMs are terrible at ARC-AGI-3