Z.ai

Numbers we checked

Each links to its source.

• glm-5-3 Benchmark score: StrongREJECT refusal rate, abliterated · measured by others · 26 Oct 12%
• glm-5-3 Benchmark score: HarmBench refusal rate, abliterated · measured by others · 26 Oct 2%
• glm-5-3 Benchmark score: JailbreakBench refusal rate, abliterated · measured by others · 26 Oct 3%
• glm-5-3 Benchmark score: Binary Exploitation (Anthropic internal, 100 tasks) · measured by others · 1 Oct 4%