Tag
Fable achieves top rank on KernelBench-Mega by writing a highly efficient CUDA megakernel with an 18.71X speedup, signaling progress toward recursive self-improvement in AI R&D.