**Abliterated model large v2: GLM 5.3 84.5% CyberGym** > GLM 5.3, abliterated f

**Abliterated model large v2: GLM 5.3 84.5% CyberGym**

> GLM 5.3, abliterated for cyber, red team, and agent testing. 84.5% CyberGym pass@1. 41.8% Terminal-Bench 4.0. 105 ExploitGym tasks in 2 hours. $5 per 1M tokens. Benchmark suite comparison Benchmarking: Abliterated Large v2 GLM 5.3 FP8 Terminal-Bench 4.0 Resolution rate 51.8% Opus 5 44.5% Fable 5 41.8% GLM 5.3 37.3% GPT-5.6 Sol CyberGym Pass@1 score 85.6% GPT-5.5 84.5% GLM 5.3 83.3% DeepSeek V4 83.1% Mythos ExploitGym...

**来源信息**
- **来源**：Hacker News：AI 热帖
- **分类**：论文
- **发布时间**：2026-09-01 18:02（北京时间）
- **原文**：[打开原文](https://abliteration.ai/blog/introducing-abliterated-model-large-v2)