GLM-5.3 · ExploitBench(GLM-5.3模型卡;unknown) 54.4 · 主文证据(ev-gmc-glm-5-3-85)

GLM-5.3 的 ExploitBench · GLM-5.3模型卡;unknown 官方出处。GLM-5.3 · max · 54.4。见《GLM-5.3 official model card》Benchmark table;ExploitBench;GLM-5.3目标列。证据 ev-gmc-glm-5-3-85。

原文与定位

GLM-5.3 · ExploitBench(GLM-5.3模型卡;unknown),GLM-5.3 · max配置,coverage-score 54.4(具体成绩公布时间未确认)。见《GLM-5.3 official model card》Benchmark table;ExploitBench;GLM-5.3目标列。 查看成绩条件

| Benchmark | GLM-5.3 | GLM-5.2 | Kimi K3 | DeepSeek-V4 Pro-0813 | Qwen3.8-Max | Opus 4.8 | Fable 5 (w/ fallback) | GPT-5.6 Sol | | ExploitBench | 54.4 | 24.4 | 32.2 | – | 28.8 | 40.0 | **78.0** | 76.5 | - **ExploitBench**: We evaluate GLM-5.3 in Claude Code 2.1.207 (max reasoning effort, no web tools with `temperature=1.0`, `top_p=1.0`, `max_new_tokens=128000`). Following the official evaluation settings, we limit the maximum number of interaction rounds between the agent and the environment to 300, and compute the average coverage score over all 41 tasks across 3 revisions. The coverage result of a task is determined by taking the union of capabilities achieved across all revisions, and the average score is obtained by averaging the results. We also apply a domain whitelist (allowing only essential domains such as pypi.org and deb.debian.org for basic tool installation) to prevent the agent from cheating. GLM-5.3 supports controlling the thinking budget through the `reasoning_effort` parameter, which accepts three levels: `low`, `high`, and `max`. It defaults to `max` if not passed (or if set to any other value). To use `low` or `high`, pass them explicitly. For benchmark and leaderboard reproduction, keep the default `max`.
官方材料
GLM-5.3 official model card
定位
主文 · Benchmark table;ExploitBench;GLM-5.3目标列
材料版本
本地2026-11-17官方卡片快照;2026-11-18完整目标列、内嵌图和原文脚注核对完成。
版本公布
未确认
披露时间
未确认 · 首发/后补未确认
采集时间
历史依据
暂无可靠历史版本依据
引用文本

GLM-5.3 · ExploitBench(GLM-5.3模型卡;unknown),GLM-5.3 · max配置,coverage-score 54.4(具体成绩公布时间未确认)。官方出处 GLM-5.3 official model card(主文,本地2026-11-17官方卡片快照;2026-11-18完整目标列、内嵌图和原文脚注核对完成。)Benchmark table;ExploitBench;GLM-5.3目标列,https://huggingface.co/zai-org/GLM-5.3/blob/main/README.md。Benchlog https://benchlog.lekshome.cc/evidence/glm-5.3/ev-gmc-glm-5-3-85/

关联成绩与条件

GLM-5.3 · ExploitBench 成绩与配置
指标与成绩配置与评测条件证据关系
coverage-score54.4原文 54.4 · 单位 未标注
公布 未确认
GLM-5.3 · max
设置
官方模型卡目标列;逐行scaffold、工具、采样及预算条件见protocol。来源把benchmark复现明确绑定max或行内明确reasoning_effort=max。
未知项
未逐项说明的指标单位、题集修订、服务精度和运行次数保持未知。
工具
允许工具
推理预算
max · 输出上限 128000 tokens
框架
Claude Code 2.1.207
协议
The card says reasoning_effort defaults to max and recommends keeping that default for benchmark reproduction. **ExploitBench**: We evaluate GLM-5.3 in Claude Code 2.1.207 (max reasoning effort, no web tools with `temperature=1.0`, `top_p=1.0`, `max_new_tokens=128000`). Following the official evaluation settings, we limit the maximum number of interaction rounds between the agent and the environment to 300, and compute the average coverage score over all 41 tasks across 3 revisions. The coverage result of a task is determined by taking the union of capabilities achieved across all revisions, and the average score is obtained by averaging the results. We also apply a domain whitelist (allowing only essential domains such as pypi.org and deb.debian.org for basic tool installation) to prevent the agent from cheating.
条件
GLM-5.3目标列;The card says reasoning_effort defaults to max and recommends keeping that default for benchmark reproduction. **ExploitBench**: We evaluate GLM-5.3 in Claude Code 2.1.207 (max reasoning effort, no web tools with `temperature=1.0`, `top_p=1.0`, `max_new_tokens=128000`). Following the official evaluation settings, we limit the maximum number of interaction rounds between the agent and the environment to 300, and compute the average coverage score over all 41 tasks across 3 revisions. The coverage result of a task is determined by taking the union of capabilities achieved across all revisions, and the average score is obtained by averaging the results. We also apply a domain whitelist (allowing only essential domains such as pypi.org and deb.debian.org for basic tool installation) to prevent the agent from cheating. 安全、内部子项或跨测试聚合仅保留原始披露,不进入常规展示。

直接支持该成绩

在排名中查看此成绩