{"version":"2026-09-27.1","checkedAt":"2026-09-27","automation":{"id":"automation","title":"AutomationBench","url":"https://zapier.com/benchmarks","repository":"https://github.com/zapier/AutomationBench","paper":"https://arxiv.org/abs/2604.18934","version":"1.0.6","checkedAt":"2026-09-27","publishedAt":null,"metric":"task_completed_correctly","scope":"官方私有留出评测集；不同方向使用各自任务","scoring":"全部计分断言通过，整题才计为成功；检查最终业务状态","configuration":"API模式，最多50步，无人在环；按各方向的最优模型与推理档位展示","limitations":"模拟业务环境；各柱可能对应不同模型，不是同一Agent的全岗表现。公开600题不是官方榜单的分母。","domains":[{"id":"sales","label":"销售","value":45.3,"model":"Claude Opus 5.5","effort":"Max + default fallbacks","second":40.17,"secondModel":"GPT-6 Astra","secondEffort":"Max","task":"CRM、线索和跨应用流程"},{"id":"marketing","label":"营销","value":50,"model":"GPT-6 Astra","effort":"Max","second":48,"secondModel":"Claude Opus 5.5","secondEffort":"Max + default fallbacks","task":"营销活动、广告分析和内容运营"},{"id":"operations","label":"运营","value":65,"model":"Claude Opus 5.5","effort":"Max + default fallbacks","second":62,"secondModel":"GPT-6 Astra","secondEffort":"Max","task":"项目、供应商和业务流程"},{"id":"support","label":"客服","value":39,"model":"GPT-6 Astra","effort":"XHigh","second":38,"secondModel":"Claude Opus 5.5","secondEffort":"Max + default fallbacks","task":"工单、SLA与知识库"},{"id":"finance","label":"财务","value":43.33,"model":"GPT-6 Astra","effort":"High","second":37.5,"secondModel":"Claude Opus 5.5","secondEffort":"Max + default fallbacks","task":"应收应付、费用和报表"},{"id":"hr","label":"HR","value":26.67,"model":"Gemini 3.8 Flash","effort":"Medium","second":25,"secondModel":"Claude Opus 5.5","secondEffort":"Max + default fallbacks","task":"招聘、入职与薪酬流程"}],"overall":[{"model":"Claude Opus 5.5","effort":"Max + default fallbacks","value":42.47},{"model":"GPT-6 Astra","effort":"Max","value":41.4},{"model":"GPT-6 Sol","effort":"XHigh","value":33.2}],"crossCheck":{"url":"https://openai.com/index/introducing-gpt-6-sol-and-luna/","date":"2026-09-22","matchedModel":"GPT-6 Sol (xhigh)","value":33.2,"note":"OpenAI发布表与Zapier榜单数值一致；属于同一评测的引用核对，不是两次独立复现。"},"fallbackNote":"Claude Opus 5.5的default fallbacks会把被拒绝的步骤交给供应商默认后备模型","relatedRoles":["FDE","GROW","DATA"]},"anthropic":{"id":"anthropic","title":"Anthropic Economic Index: Cadences","url":"https://www.anthropic.com/research/economic-index-june-2026-report","dataset":"https://economic-research.anthropic.com/releases/econ-index/release-2026-06-26.zip","version":"release-2026-06-26","publishedAt":"2026-06-26","checkedAt":"2026-09-27","periodStart":"2026-05-01","periodEndExclusive":"2026-06-01","license":"CC-BY","archiveSha256":"9be20dd451b91e4d59bd605528a8d5e55f6a182c1d280172f303f871553236da","metric":"collaboration_bucket_automation_pct","scope":"GLOBAL / soc_occupation / hierarchy_level=1","limitations":"职业相关使用中的交互分类；不是该职业所有任务的自动化率或验收成功率。1P API排除Claude Code；聊天样本含Chat与Cowork。","rows":[{"id":"15","label":"计算机与数学","chat":63.9,"api":94.58,"chatAugmentation":36.1,"apiAugmentation":5.42},{"id":"13","label":"商业与财务","chat":42.61,"api":94.92,"chatAugmentation":57.39,"apiAugmentation":5.08},{"id":"11","label":"管理","chat":43.9,"api":95.58,"chatAugmentation":56.1,"apiAugmentation":4.42},{"id":"27","label":"艺术与传媒","chat":35.85,"api":94.83,"chatAugmentation":64.15,"apiAugmentation":5.17},{"id":"41","label":"销售","chat":43.79,"api":90.78,"chatAugmentation":56.21,"apiAugmentation":9.22},{"id":"43","label":"办公与行政","chat":56.02,"api":94.41,"chatAugmentation":43.98,"apiAugmentation":5.59}],"sourceFiles":["aei_claude_ai_2026-06-26.csv","aei_1p_api_2026-06-26.csv"],"metrics":["collaboration_bucket_automation_pct","collaboration_bucket_augmentation_pct"]},"gdpval":{"id":"gdpval","title":"GDPval：专业交付物质量","url":"https://openai.com/index/introducing-gpt-5-5/","methodUrl":"https://openai.com/index/gdpval/","publishedAt":"2026-04-23","checkedAt":"2026-09-27","metric":"wins or ties","scope":"44个职业的明确知识工作任务；与专业人员产物比较","configuration":"OpenAI发布表 Professional / GDPval；保留发布时配置与口径","limitations":"专家偏好比较的胜或平，不是业务流程严格通过率。此处为4月历史结果，不声称当前SOTA。","rows":[{"model":"GPT-5.5","value":84.9},{"model":"GPT-5.4","value":83},{"model":"Claude Opus 4.7","value":80.3}]},"reviewNote":"本次查阅Anthropic六月Economic Index、三月任务数据，OpenAI四月GDPval及九月模型报告，并沿引用核对Zapier官方结果和评分仓库。六月经济数据是本轮找到的最近公开发布，九月OpenAI报告没有GDPval新表。"}