Hourglass Reasoning

Graph relationships

1. USES
   GPT-5.5

   Hourglass Reasoning achieves 58% Verilog synthesis accuracy with GPT-5.5.

   Source finding
2. BENCHMARKED AGAINST
   ChipBench

   Hourglass Reasoning lifts ChipBench Verilog synthesis accuracy from 31% to 58%.

   Source finding
3. BENCHMARKED AGAINST
   ARC-AGI-2

   Hourglass Reasoning improves ARC-AGI-2 best-of-5 accuracy by 14 points.

   Source finding

Source trail

Graph sources