Home Model series GLM GLM-5 GLM-5 · Slide generation human evaluation · GLM-5固定v2报告/模型卡;版本未注明 · Scores and configurations GLM-5 · Slide generation human evaluation · GLM-5固定v2报告/模型卡;版本未注明 · Scores and configurations GLM-5 record for Slide generation human evaluation · GLM-5固定v2报告/模型卡;版本未注明. 4 score records for this release, including 4 exact values (may include historical records). Includes 4 source locations.
Scores and evaluation conditions GLM-5 · Slide generation human evaluation scores and configurations Metric and score Configuration and evaluation conditions Sources Content quality win rate 60% Reported 60% · Unit % Published Unconfirmed
GLM-5 · 报告与固定模型卡 Settings 最终后训练GLM-5;各行scaffold、解码与工具条件分别见protocol,未将工具或来源另立为推理档位。 Unknowns 来源未逐项明确思考开关、reasoning effort、服务精度与完整题集修订。、Tool-Decathlon报告39.2与固定卡38.0均保留;BrowseComp上下文管理的正文与附录方法描述不一致。 Tools Tool access not recorded Reasoning budget not recorded Framework not recorded Protocol Human evaluation of GLM-5 versus GLM-4.5 after the slide-generation RL, rejection sampling, and masking refinement pipeline. Conditions Human evaluation of GLM-5 versus GLM-4.5 after the slide-generation RL, rejection sampling, and masking refinement pipeline. View this score in rankings Layout rationality win rate 57.5% Reported 57.5% · Unit % Published Unconfirmed
GLM-5 · 报告与固定模型卡 Settings 最终后训练GLM-5;各行scaffold、解码与工具条件分别见protocol,未将工具或来源另立为推理档位。 Unknowns 来源未逐项明确思考开关、reasoning effort、服务精度与完整题集修订。、Tool-Decathlon报告39.2与固定卡38.0均保留;BrowseComp上下文管理的正文与附录方法描述不一致。 Tools Tool access not recorded Reasoning budget not recorded Framework not recorded Protocol Human evaluation of GLM-5 versus GLM-4.5 after the slide-generation RL, rejection sampling, and masking refinement pipeline. Conditions Human evaluation of GLM-5 versus GLM-4.5 after the slide-generation RL, rejection sampling, and masking refinement pipeline. View this score in rankings Visual aesthetics win rate 65% Reported 65% · Unit % Published Unconfirmed
GLM-5 · 报告与固定模型卡 Settings 最终后训练GLM-5;各行scaffold、解码与工具条件分别见protocol,未将工具或来源另立为推理档位。 Unknowns 来源未逐项明确思考开关、reasoning effort、服务精度与完整题集修订。、Tool-Decathlon报告39.2与固定卡38.0均保留;BrowseComp上下文管理的正文与附录方法描述不一致。 Tools Tool access not recorded Reasoning budget not recorded Framework not recorded Protocol Human evaluation of GLM-5 versus GLM-4.5 after the slide-generation RL, rejection sampling, and masking refinement pipeline. Conditions Human evaluation of GLM-5 versus GLM-4.5 after the slide-generation RL, rejection sampling, and masking refinement pipeline. View this score in rankings Overall win rate 67.5% Reported 67.5% · Unit % Published Unconfirmed
GLM-5 · 报告与固定模型卡 Settings 最终后训练GLM-5;各行scaffold、解码与工具条件分别见protocol,未将工具或来源另立为推理档位。 Unknowns 来源未逐项明确思考开关、reasoning effort、服务精度与完整题集修订。、Tool-Decathlon报告39.2与固定卡38.0均保留;BrowseComp上下文管理的正文与附录方法描述不一致。 Tools Tool access not recorded Reasoning budget not recorded Framework not recorded Protocol Human evaluation of GLM-5 versus GLM-4.5 after the slide-generation RL, rejection sampling, and masking refinement pipeline. Conditions Human evaluation of GLM-5 versus GLM-4.5 after the slide-generation RL, rejection sampling, and masking refinement pipeline. View this score in rankings
Benchmark versions and metrics Aliases Slide generation human evaluation
Content quality win rate · % · Higher is better Layout rationality win rate · % · Higher is better Visual aesthetics win rate · % · Higher is better Overall win rate · % · Higher is better