Tencent's Hy4 Tops CWE-Bench-AA at 76%

Key Info

Tencent's Hy4 model ranks #1 on CWE-Bench-AA with a score of 76%, six points ahead of DeepSeek V4.1 Flash 1017 (max).

Highlights

• Hy4 achieves 76% on CWE-Bench-AA, placing it first on the leaderboard
• The margin over DeepSeek V4.1 Flash 1017 (max) is 6 percentage points
• CWE-Bench-AA is a benchmark tied to software weakness (CWE) detection tasks