ModelsSelected

KernelZero-7B beats Claude and DeepSeek on GPU kernel generation

KernelZero: Co-Evolving Proposer and Coder for Continuously Improved GPU Kernel Generation

Why we picked it

A 7B open model trained with an automatic curriculum now beats frontier LLMs on kernel generation, suggesting specialised small models can outperform general giants on narrow engineering tasks. Builders writing CUDA or Triton kernels can watch for KernelZero's release to automate optimisation of Torch modules. The comparison against DeepSeek-V4-Pro on Triton gives a concrete benchmark for teams weighing frontier APIs against compact task-specific models.

KernelZero is a co-evolution framework pairing a Proposer that generates Torch modules with a Coder that translates them into CUDA or Triton kernels.

Body

It uses frontier-driven module generation to produce training data aligned with the Coder's current weaknesses, plus Correctness-Aware Group Relative Policy Optimization (CA-GRPO) that optimizes performance only after correctness is reliable.

KernelZero-7B surpasses Claude-4.5-Sonnet on CUDA and DeepSeek-V4-Pro on Triton.

On KernelBench Level 1 and 2 it reaches CUDA pass@1 scores of 75.8% and 69.

6%, with pass@10 of 100% and 97%, and Triton pass@1 scores of 77.2% and 72.

5%.

Original · arXiv: DeepSeek

KernelZero: Co-Evolving Proposer and Coder for Continuously Improved GPU Kernel Generation

High-performance GPU kernels are essential to modern machine learning systems, yet automatically generating kernels that are both correct and efficient remains challenging. Existing LLM-based approaches face two major limitations: the scarcity of high-quality training data aligned with the model's current capabilities, and the inherent trade-off between kernel correctness and performance. To address these challenges, we propose KernelZero, a co-evolution framework that continuously improves GPU kernel generation through two specialized models: a Proposer that generates Torch modules from API sets and a Coder that translates them into CUDA or Triton kernels. KernelZero uses a frontier-driven module generation mechanism to continuously produce capability-aligned training modules based on the Coder's current weaknesses. It further introduces Correctness-Aware Group Relative Policy Optimization (CA-GRPO), which optimizes performance only after correctness becomes sufficiently reliable. By alternating the optimization of the Proposer and Coder, KernelZero forms an automatic curriculum that enables targeted and training-efficient capability improvement. Empirically, KernelZero-7B surpasses Claude-4.5-Sonnet on CUDA and DeepSeek-V4-Pro on Triton. On KernelBench Level 1 and 2, it achieves CUDA pass@1 scores of 75.8% and 69.6%, respectively, with pass@10 reaching 100% and 97%. On Triton, it achieves pass@1 scores of 77.2% and 72.5%, respectively.

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include KernelZero, CUDA, Triton, KernelBench.

Primary source verification: published directly by arXiv: DeepSeek with first-party authority.