Kernel Arena - KernelBench HIP: leaderboard
Kernel Arena: mean correctness plus speedup over 41 HIP kernel tasks on AMD MI300X, using the unsuffixed basic-harness source runs. The separately published _min run group is retained on its own board; upstream does not define that suffix. Scores come from Wafer’s official suite exports.
Metric: Mean Correctness+Speedup. Source: github.com. Saturation forecast: Around April 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 15.46 | #110 |
| 2 | Gemini 3 Pro | 4.21 | #80 |
| 3 | Claude Opus 4.5 | 3.11 | #83 |
| 4 | Gemini 3 Flash | 2.51 | #99 |
| 5 | Grok 4.1 | 2.02 | #227 |
| 6 | Kimi K2 | 1.5 | #243 |
| 7 | GLM-4.7 | 1.31 | #192 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kernel-arena-kernelbench-hip · How It Works · Data refreshed daily, snapshot 2026-10-20.