Doubao Opens Third-Party Model Selection and Faces WorkBuddy Test
A Juejin author reports that Doubao began letting users select third-party models on November 12 and observed GLM-5.3 Flash in its model list. Using the same model, prompt, and classroom random-roll-call webpage task, Doubao took 19 seconds to respond and 1 minute 33 seconds to finish, versus about 10 seconds and 1 minute 16 seconds for WorkBuddy. The author opened both outputs in a browser and tested scrolling, stopping, answer-card display, and other interactions, finding that both produced functional webpages.
The report provides a controlled, first-hand comparison of two AI tools running the same model and task, while highlighting a potentially broader product option in Doubao.
一次不掺水的同模型横向实测:豆包新放开的多模型选择,能否作为 WorkBuddy 的备选?控制变量做得很干净——同一个模型(GLM-5.3 Flash)、同一份提示词(一个字没改)、同一道题(做一个课堂随机点名器网页),只换承载工具。几个工程视角值得看:一是豆包侧多出「推理强度」三档(低/高/最高),WorkBuddy 默认跑、无此选项;二是豆包首响慢(19 秒 vs 10 秒),但会主动补做未要求的功能(点过去重、空格键触发、名单数组化);三是两边产物都做了可交互性验证,不是只看「生成完成」——point 开浏览器点按钮、名单滚动、停住、弹卡全流程走通。本文按六步还原,附完整提示词、耗时对比与「备选工具」的分场景判断。
我是小虎,浙大计算机本硕,破局行动家,《AI超级个体——一人公司完全指南》作者,专注于AI培训、超级个体打造、公司提效陪跑。