[{"data":1,"prerenderedAt":80},["ShallowReactive",2],{"tool-mmbench":3},{"tool":4,"relatedTools":42},{"id":5,"name":6,"slug":7,"tagline":8,"description":9,"websiteUrl":10,"logoText":11,"pricing":12,"editorScore":13,"clickCount":14,"featured":15,"latest":15,"updatedAt":16,"category":17,"tags":26,"overview":28,"highlights":29,"useCases":38},"dba9d1e924df979e2f7f7fc5cbf819ae","MMBench","mmbench","全方位的多模态大模型能力评测体系","MMBench是多模态基准测试，由上海人工智能实验室、南洋理工大学、香港中文大学、新加坡国立大学和浙江大学的研究人员联合推出。MMBench推出一个综合评估流程，从感知到认知能力逐级细分评估，覆盖20项细粒度能力，从互联网与权威基准数据集采集约3000道单项选择题。","https:\u002F\u002Fmmbench.opencompass.org.cn\u002F?utm_source=ai-bot.cn","MMB","官网可查看",98,3,false,"2026-04-15T04:26:34+08:00",{"id":18,"name":19,"slug":20,"description":21,"iconLabel":22,"accentColor":23,"toolCount":24,"sortOrder":25},"d6d303a9f41677e5f1a5f9b12c520225","智能体与自动化","agents","适合搭建业务工作流、插件编排和多步骤自动化，用来把 AI 接进真实流程。","AUTO","#6C7A2D",216,7,[27],"AI模型评测","MMBench是多模态基准测试，由上海人工智能实验室、南洋理工大学、香港中文大学、新加坡国立大学和浙江大学的研究人员联合推出。MMBench推出一个综合评估流程，从感知到认知能力逐级细分评估，覆盖20项细粒度能力，从互联网与权威基准数据集采集约3000道单项选择题。打破常规一问一答基于规则匹配提取选项进行评测，循环打乱选项验证输出结果的一致性，基于ChatGPT精准匹配模型回复至选项。MMBench涵盖多种任务类型，如视觉问答、图像描述生成等，基于综合多维度指标，为模型提供全面的性能评估。MMBench 的排行榜展示不同模型在这些任务上的表现，帮助研究者和开发者了解当前多模态技术的发展水平，推动相关领域的技术进步。 细粒度能力评估 ：将多模态能力细分为多个维度（如感知、推理等），针对每个维度设计相关问题，全面评估模型的细粒度能力。",[30,31,32,33,34,35,36,37],"细粒度能力评估","大规模多模态数据集","创新评估策略","多语言支持","数据可视化","官方评估工具","基准测试与排行榜","安装依赖",[39,40,41],"多种任务类型","图像描述生成等","约 3000 个多项选择题",[43,56,67],{"id":44,"name":45,"slug":46,"tagline":47,"description":48,"websiteUrl":49,"logoText":50,"pricing":12,"editorScore":13,"clickCount":51,"featured":52,"latest":52,"updatedAt":16,"category":53,"tags":54},"a638bfd4cc081d0294d7d12975a727dd","GPT-4","gpt-4","OpenAI旗下最新的GPT-4模型","GPT-4（Generative Pre-trained Transformer）是 OpenAI 开发的自然语言处理模型 GPT 家族中的第四个版本，该模型依靠强大的神经网络来理解和生成类似人类的语言。","https:\u002F\u002Fcn.bing.com\u002Fsearch?q=GPT-4&ensearch=1","GPT",2356,true,{"id":18,"name":19,"slug":20,"description":21,"iconLabel":22,"accentColor":23,"toolCount":24,"sortOrder":25},[55],"AI训练模型",{"id":57,"name":58,"slug":59,"tagline":60,"description":61,"websiteUrl":62,"logoText":63,"pricing":12,"editorScore":13,"clickCount":64,"featured":52,"latest":52,"updatedAt":16,"category":65,"tags":66},"a23e89d9e4203bd1c683c41e837f370e","LMArena","lmarena","AI模型评估平台","LMArena是加州大学伯克利分校推出的创新AI模型评估平台，基于让用户对不同AI模型的回答进行匿名投票，衡量模型的表现。用户输入问题后，平台提供两个模型的回答，用户根据偏好选择更优答案，投票结果直接塑造公共排行榜。","https:\u002F\u002Flmarena.ai\u002F?utm_source=ai-bot.cn","LMA",681,{"id":18,"name":19,"slug":20,"description":21,"iconLabel":22,"accentColor":23,"toolCount":24,"sortOrder":25},[27],{"id":68,"name":69,"slug":70,"tagline":71,"description":72,"websiteUrl":73,"logoText":74,"pricing":75,"editorScore":13,"clickCount":76,"featured":52,"latest":52,"updatedAt":16,"category":77,"tags":78},"32bb7d109995125694dc4a175a840117","Coze","coze","海量AI智能体免费用，一键复制同款","Coze是字节跳动推出的AI办公助手，能通过AI技术简化工作流程，提升工作效率。通过Agent Skills技能系统和Agent Plan长期计划功能，实现从单一工具到智能工作伙伴的转变。","https:\u002F\u002Fwww.coze.cn\u002Foverview?utm_medium=daohang&utm_source=aibot&utm_term=49156769m","COZ","免费",316,{"id":18,"name":19,"slug":20,"description":21,"iconLabel":22,"accentColor":23,"toolCount":24,"sortOrder":25},[79],"AI开发平台",1776262185604]