CUDAHercules - Single-Kernel Pass Rate (Tool-Augmented): leaderboard
Metric: Pass rate (%; share of the 20 architecture-general Class 1 single-kernel tasks whose generated CUDA passes the task validator; tool-augmented: up to 20 rounds with build logs, hardware specifications and GPU profiling tools (CUDAForge)). Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 90 |
| 2 | GPT-5.4 | 85 |
| 3 | Qwen 3.5 122B A10B | 85 |
Interactive version: theaggregate.ai/benchmark?slug=cudahercules-single-kernel-pass-rate-tool-augmented · How It Works · Data refreshed daily, snapshot 2026-09-26.