白泽（Whitzard）女娲实验室 ↗
NUWANUWA INDEX

搜索 NUWA Index

研究 · 测量协议 · 评测结果 · 观点

English

MODEL COMPARISON

比较模型与智能体系统

选择风险主题与测量协议，对照 2–4 个受测系统。任务能力、行为倾向与观察行为保留各自含义。

CyberGym · 101 项共同困难任务 · 2M tokens

SaferAI 的 GLM-5.2 实验与转引的 Lyptus GPT-5.5、Opus 4.6 运行，均限制在相同 101 项任务。

描述性比较：脚手架、提供方和推理条件存在差异，当前结果不构成严格模型排名。

GLM-5.2 (GLM-5.2); CyberGym pass@1: 36.63%; 37/101; CyberGym Level 1 hard-task matched subset; {"modelVersion":null,"harness":"Inspect Evals","tools":null,"budget":"2M tokens per task"}; ? attempts; evaluator SaferAI; reporter SaferAI; https://www.safer-ai.org/u/2026/08/SaferAI-GLM-Evaluation-Report.pdf; v1.1 · revised 26 August 2026; p.8 Fig.3 printed bar label; released 2026-06-16; evaluated ?; published 2026-08-02; public 2026-08-26; ; 来源转录不等于实验复现。未确定的日期、执行方及配置保留为未知；原文分组不作为经核验排名。 本条限制说明（英文，完整保留）：122 GLM tasks run; comparison restricted to 101 matched tasks. Nineteen tasks lacked comparator coverage; two tasks excluded as defective. Hard-task selection not representative of all 1507 tasks. GPT-5.5 (GPT-5.5); CyberGym pass@1: 55.45%; 56/101; CyberGym Level 1 hard-task matched subset; {"modelVersion":null,"harness":null,"tools":null,"budget":"2M tokens per task"}; ? attempts; evaluator Lyptus Research; reporter SaferAI; https://www.safer-ai.org/u/2026/08/SaferAI-GLM-Evaluation-Report.pdf; v1.1 · revised 26 August 2026; p.8 Fig.3 printed bar label; released 2026-04-23; evaluated ?; published 2026-08-02; public 2026-08-26; ; 来源转录不等于实验复现。未确定的日期、执行方及配置保留为未知；原文分组不作为经核验排名。 本条限制说明（英文，完整保留）：122 GLM tasks run; comparison restricted to 101 matched tasks. Nineteen tasks lacked comparator coverage; two tasks excluded as defective. Hard-task selection not representative of all 1507 tasks. Claude Opus 4.6 (Claude Opus 4.6); CyberGym pass@1: 44.55%; 45/101; CyberGym Level 1 hard-task matched subset; {"modelVersion":null,"harness":null,"tools":null,"budget":"2M tokens per task"}; ? attempts; evaluator Lyptus Research; reporter SaferAI; https://www.safer-ai.org/u/2026/08/SaferAI-GLM-Evaluation-Report.pdf; v1.1 · revised 26 August 2026; p.8 Fig.3 printed bar label; released 2026-02-05; evaluated ?; published 2026-08-02; public 2026-08-26; ; 来源转录不等于实验复现。未确定的日期、执行方及配置保留为未知；原文分组不作为经核验排名。 本条限制说明（英文，完整保留）：122 GLM tasks run; comparison restricted to 101 matched tasks. Nineteen tasks lacked comparator coverage; two tasks excluded as defective. Hard-task selection not representative of all 1507 tasks.CyberGym · 101 项共同困难任务 · 2M tokensCyberGym pass@1 · % · descriptiveSaferAI 的 GLM-5.2 实验与转引的 Lyptus GPT-5.5、Opus 4.6 运行，均限制在相同 101 项任务。0255075100GLM-5.2; 36.63%; SaferAIGLM-5.236.63%SaferAIGPT-5.5; 55.45%; Lyptus ResearchGPT-5.555.45%Lyptus ResearchClaude Opus 4.6; 44.55%; Lyptus ResearchClaude Opus 4.644.55%Lyptus ResearchNUWA：来源汇编；非 NUWA 实验。完整名称、条件与出处见数据表。Snapshot: 2026-10-29.1; GLM-5.2 Risk Evaluation Report; GLM-5.2 (GLM-5.2); CyberGym pass@1: 36.63%; 37/101; CyberGym Level 1 hard-task matched subset; {"modelVersion":null,"harness":"Inspect Evals","tools":null,"budget":"2M tokens per task"}; ? attempts; evaluator SaferAI; reporter SaferAI; https://www.safer-ai.org/u/2026/08/SaferAI-GLM-Evaluation-Report.pdf; v1.1 · revised 26 August 2026; p.8 Fig.3 printed bar label; released 2026-06-16; evaluated ?; published 2026-08-02; public 2026-08-26; ; 来源转录不等于实验复现。未确定的日期、执行方及配置保留为未知；原文分组不作为经核验排名。 本条限制说明（英文，完整保留）：122 GLM tasks run; comparison restricted to 101 matched tasks. Nineteen tasks lacked comparator coverage; two tasks excluded as defective. Hard-task selection not representative of all 1507 tasks. GPT-5.5 (GPT-5.5); CyberGym pass@1: 55.45%; 56/101; CyberGym Level 1 hard-task matched subset; {"modelVersion":null,"harness":null,"tools":null,"budget":"2M tokens per task"}; ? attempts; evaluator Lyptus Research; reporter SaferAI; https://www.safer-ai.org/u/2026/08/SaferAI-GLM-Evaluation-Report.pdf; v1.1 · revised 26 August 2026; p.8 Fig.3 printed bar label; released 2026-04-23; evaluated ?; published 2026-08-02; public 2026-08-26; ; 来源转录不等于实验复现。未确定的日期、执行方及配置保留为未知；原文分组不作为经核验排名。 本条限制说明（英文，完整保留）：122 GLM tasks run; comparison restricted to 101 matched tasks. Nineteen tasks lacked comparator coverage; two tasks excluded as defective. Hard-task selection not representative of all 1507 tasks. Claude Opus 4.6 (Claude Opus 4.6); CyberGym pass@1: 44.55%; 45/101; CyberGym Level 1 hard-task matched subset; {"modelVersion":null,"harness":null,"tools":null,"budget":"2M tokens per task"}; ? attempts; evaluator Lyptus Research; reporter SaferAI; https://www.safer-ai.org/u/2026/08/SaferAI-GLM-Evaluation-Report.pdf; v1.1 · revised 26 August 2026; p.8 Fig.3 printed bar label; released 2026-02-05; evaluated ?; published 2026-08-02; public 2026-08-26; ; 来源转录不等于实验复现。未确定的日期、执行方及配置保留为未知；原文分组不作为经核验排名。 本条限制说明（英文，完整保留）：122 GLM tasks run; comparison restricted to 101 matched tasks. Nineteen tasks lacked comparator coverage; two tasks excluded as defective. Hard-task selection not representative of all 1507 tasks.
系统／模型结果／样本运行配置归属／日期
GLM-5.2capability36.63%—37/101modelVersion: —harness: Inspect Evalstools: —budget: 2M tokens per taskCyberGym Level 1 hard-task matched subset评测方: SaferAISaferAI评测 — · 报告 2026-08-02证据公开 2026-08-26
GPT-5.5capability55.45%—56/101modelVersion: —harness: —tools: —budget: 2M tokens per taskCyberGym Level 1 hard-task matched subset评测方: Lyptus ResearchSaferAI评测 — · 报告 2026-08-02证据公开 2026-08-26
Claude Opus 4.6capability44.55%—45/101modelVersion: —harness: —tools: —budget: 2M tokens per taskCyberGym Level 1 hard-task matched subset评测方: Lyptus ResearchSaferAI评测 — · 报告 2026-08-02证据公开 2026-08-26
比较条件与解释范围
  • Task intersection · matched
    101 项共同 Level1 困难任务；122 项初始任务扣除比较模型未覆盖的 19 项及两项有缺陷任务。
  • Metric / budget · matched
    每任务成功、pass@1、2M token 设置；本图不合并高预算重试叠加。
  • Evaluator / original runs · documented
    GLM-5.2：SaferAI。GPT-5.5、Opus4.6：Lyptus；SaferAI 转引这些运行。原始聚合数据可核对 101 项任务子集计数。
  • Harness / provider / reasoning · different
    Inspect Evals 与 Lyptus ReAct 实现、端点提供方、推理与输出限制存在差异或尚未完整匹配，不能将差异完全归因于基础模型。

来源报告 95% Wilson 区间存在重叠；图中未印出精确端点，因此不转录端点。本图是任务匹配的描述性比较，不构成严格条件匹配的模型排名。

CSV ↓SVG ↓比较协议设计 →

证据快照 2026-10-29.1 · NUWA Index 1.0.0