GLM-5.3 · SWE-Marathon (v1.1)(GLM-5.3模型卡;v1.1) 42.5 · 主文证据(ev-glm53-swe-marathon)

GLM-5.3 的 SWE-Marathon (v1.1) · GLM-5.3模型卡;v1.1 官方出处。GLM-5.3 · max · 42.5。见《GLM-5.3 model card | zai-org》Benchmark table, SWE-Marathon (v1.1) row, GLM-5.3 column; SWE-Marathon footnote。证据 ev-glm53-swe-marathon。

原文与定位

GLM-5.3 · SWE-Marathon (v1.1)(GLM-5.3模型卡;v1.1),GLM-5.3 · max配置,Reported score 42.5(具体成绩公布时间未确认)。见《GLM-5.3 model card | zai-org》Benchmark table, SWE-Marathon (v1.1) row, GLM-5.3 column; SWE-Marathon footnote。 查看成绩条件

| SWE-Marathon (v1.1) | 42.5 | 19.4 | 48.1 | – | – | **48.8** | 33.1 | 42.5 | **SWE-Marathon**: We evaluate GLM-5.3 using Claude Code 2.1.207 with maximum effort level, `temperature = 1.0`, `top_p = 0.95`, `max_new_tokens = 128000`, and a 1M-token context window. For `strip-clone`, the original anti-cheat checks used overly broad import detection that could reject valid implementations. We removed the affected checks and performed llm-based inspection instead to avoid false positives. For `parameter-golf` and `trimul-cuda`, changes to the NVIDIA wheels caused the Docker image builds to fail, so we added `--extra-index-url https://pypi.org/simple` to restore successful builds.
官方材料
GLM-5.3 model card | zai-org
定位
主文 · Benchmark table, SWE-Marathon (v1.1) row, GLM-5.3 column; SWE-Marathon footnote
材料版本
官方 Hugging Face main 分支模型卡;表格、评测脚注与配置说明已核读;2026-11-17 采集
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 · SWE-Marathon (v1.1)(GLM-5.3模型卡;v1.1),GLM-5.3 · max配置,Reported score 42.5(具体成绩公布时间未确认)。官方出处 GLM-5.3 model card | zai-org(主文,官方 Hugging Face main 分支模型卡;表格、评测脚注与配置说明已核读;2026-11-17 采集)Benchmark table, SWE-Marathon (v1.1) row, GLM-5.3 column; SWE-Marathon footnote,https://huggingface.co/zai-org/GLM-5.3/blob/main/README.md。Benchlog https://benchlog.lekshome.cc/evidence/glm-5.3/ev-glm53-swe-marathon/

关联成绩与条件

GLM-5.3 · SWE-Marathon (v1.1) 成绩与配置
指标与成绩配置与评测条件证据关系
Reported score42.5原文 42.5 · 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
工具条件未记录
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. **SWE-Marathon**: We evaluate GLM-5.3 using Claude Code 2.1.207 with maximum effort level, `temperature = 1.0`, `top_p = 0.95`, `max_new_tokens = 128000`, and a 1M-token context window. For `strip-clone`, the original anti-cheat checks used overly broad import detection that could reject valid implementations. We removed the affected checks and performed llm-based inspection instead to avoid false positives. For `parameter-golf` and `trimul-cuda`, changes to the NVIDIA wheels caused the Docker image builds to fail, so we added `--extra-index-url https://pypi.org/simple` to restore successful builds.
条件
GLM-5.3; exact score definition and unit unspecified. 2026-11-18按完整模型卡修正单位、指标/题集或协议标注,数字不变。

直接支持该成绩

在排名中查看此成绩