CUDA Agent: Large-Scale Agentic RL for High-Performance CUDA Kernel Generation

TL;DR

CUDA Agent introduces a large-scale reinforcement learning system that significantly improves CUDA kernel generation, outperforming existing models and compiler-based systems in speed and optimization quality.

Contribution

It presents a novel agentic RL framework with scalable data synthesis, automated verification, and stable training techniques for CUDA kernel optimization.

Findings

Achieves 100% faster performance than torch.compile on KernelBench levels.

Outperforms proprietary models by about 40% on the hardest benchmark level.

Demonstrates state-of-the-art results in CUDA kernel optimization.

Abstract

GPU kernel optimization is fundamental to modern deep learning but remains a highly specialized task requiring deep hardware expertise. Despite strong performance in general programming, large language models (LLMs) remain uncompetitive with compiler-based systems such as torch.compile for CUDA kernel generation. Existing CUDA code generation approaches either rely on training-free refinement or fine-tune models within fixed multi-turn execution-feedback loops, but both paradigms fail to fundamentally improve the model's intrinsic CUDA optimization ability, resulting in limited performance gains. We present CUDA Agent, a large-scale agentic reinforcement learning system that develops CUDA kernel expertise through three components: a scalable data synthesis pipeline, a skill-augmented CUDA development environment with automated verification and profiling to provide reliable reward signals, and reinforcement learning algorithmic techniques enabling stable training. CUDA Agent achieves state-of-the-art results on KernelBench, delivering 100\%, 100\%, and 92\% faster rate over torch.compile on KernelBench Level-1, Level-2, and Level-3 splits, outperforming the strongest proprietary models such as Claude Opus 4.5 and Gemini 3 Pro by about 40\% on the hardest Level-3 setting.