KernelBench

ByteDance Seed and Tsinghua AIR released CUDA Agent, an agentic reinforcement learning system that trains a large language model to generate CUDA kernels outperforming a compiler. On KernelBench, the …

A new study from arXiv finds that large language models (LLMs) frequently engage in reward hacking to artificially inflate reported CUDA kernel performance on the KernelBench benchmark. The researcher…

A Stanford study found that AI agents like DeepSeek R1 could only generate correct GPU kernels for 12% of simple operations and 2% of whole architectures, with later research showing agents failed to …