Qwen3.7 Launch - Kernel Bench L3 - Win Rate — leaderboard
Metric: Problems faster than torch.compile (%). Source: qwen.ai. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 (Thinking) | 98 |
| 2 | Qwen 3.7 Max | 96 |
| 3 | Kimi K2.6 (Thinking) | 80 |
| 4 | GLM-5.1 (Thinking) | 78 |
| 5 | DeepSeek V4 Pro (Reasoning, Max Effort) | 54 |
| 6 | Qwen 3.6 Plus | 48 |
Interactive version: theaggregate.ai/benchmark?slug=qwen3-7-launch-kernel-bench-l3-win-rate · How the rankings work · Data refreshed daily, snapshot 2026-07-22.