CANN Bench opens agent kernel generation beyond CUDA.

A 15-author Huawei team argues kernel-generation benchmarks are almost entirely CUDA and Triton, leaving less-documented hardware with no shared yardstick (arXiv 2607.20518). CANN Bench covers 53 operators and 1,060 test cases in four difficulty tiers, from elementwise primitives up to MoE dispatch and FlashAttention, across FP16, BF16, FP32, and INT8. Scoring is a three-axis weighted composite treating compilation, functional correctness, and performance independently, graded against both PyTorch-on-Ascend and an analytical per-case Hardware-Anchored Performance limit measured on real NPU silicon. Explicitly designed to resist reward hacking, versioned inside the official CANN repo.