LectureRechercheRadarCadre d'investissement
Connexion / Inscription
← Retour à la lecture
ActualitésSujet · 千问办公模型选型Matériel2 min de lectureRédaction DisconfirmAI

Alibaba's Qianwen Office Integrates Qwen4-Max and Third-Party Models, Opens User Selection

Alibaba's Qianwen Office adds 10 LLMs, including its new Qwen4-Max and competitors like Kimi and DeepSeek, allowing users to select models based on task needs and enabling enterprise management.

2026-11-16 19:44 GMT+8

Zhidx reported on November 16 that Alibaba's AI office product, "Qianwen Office," underwent a major upgrade by launching 10 large language models at once. The update includes not only Alibaba's newly released flagship model, Qwen4-Max, but also several third-party models such as Kimi-K3.1-Pro, GLM-5.3, DeepSeek-V4-Pro, MiniMax-M3.1-Flash, and Step-5-Preview. This signifies that Qianwen Office has officially opened the "model selection" feature to its users.

Previously, Qianwen Office primarily offered three fixed tiers: Standard, Advanced, and Qwen3.8-Max. With this upgrade, users can flexibly choose different models for tasks based on complexity, response speed requirements, and token cost budgets. For instance, Qwen4-Max is recommended for complex reasoning tasks; GLM-5.3 or DeepSeek-V4-Pro can be used for rule sorting, plan checking, and other analytical tasks; while lower-cost models like Qwen3.8-Flash, GLM-5.3-Flash, or DeepSeek-V4.1-Flash are suitable for high-frequency office tasks such as copywriting, information extraction, and content organization.

Notably, the newly added Qwen4-Max is the first model in Alibaba's Qwen4 series, released today. It features a new architecture with 2.4 trillion total parameters, activating 71 billion parameters per token, native visual understanding capabilities, and support for a 1 million token context window. According to evaluation results published by Alibaba, Qwen4-Max shows significant improvements over the previous Qwen3.8-Max across multiple professional Agent benchmarks. Specifically, in the Work360 benchmark, the average cost per single task dropped from 26.58 yuan to 11.16 yuan, a decrease of approximately 58%.

Connexion / Inscription
LectureRechercheRadarCadre d'investissement

In addition to individual user choices, Qianwen Office has opened model management permissions to enterprise users. Enterprise administrators can configure model usage rights based on departments, user groups, or specific members, assigning corresponding models for different roles and tasks while centrally managing internal model resources. This move reflects a pragmatic trend in the current AI office sector: given that different models have advantages in reasoning capability, speed, and cost, integrating the strengths of various providers is becoming a more effective solution than relying on a single model.

Sources

  • zhidx.com↗
Sujet

千问办公模型选型

2026年11月16日,阿里巴巴旗下AI办公产品“千问办公”正式开放模型自主选型功能,一次性上线包括自家最新旗舰Qwen4-Max及Kimi、GLM、DeepSeek等在内的10款大模型。此前该产品仅提供标准、高级和Qwen3.8-Max三…

3 étapes1 piècesMis à jour 2026-11-19 02:33

Comment on en est arrivé là

  1. 2026-11-16阿里发布新一代旗舰模型Qwen4-Max,拥有2.4万亿总参数,支持100万Token上下文窗口。zhidx.com ↗
  2. 2026-11-16千问办公上线10款大模型,除Qwen4-Max外,还引入Kimi-K3.1-Pro、GLM-5.3、DeepSeek-V4-Pro等第三方模型,正式开放用户自主选型。finance.biggo.com ↗
  3. 2026-11-16千问办公向企业用户开放模型管理权限,管理员可按部门或用户组配置模型使用权限,实现组织级治理。finance.biggo.com ↗

Ce qui reste non résolu

  • 各接入模型的具体计费标准及免费额度限制何时公布?
  • Agent如何自动识别任务类型并调用最优模型,而非仅依赖用户手动选择?
  • 多模型集成后,数据在不同厂商模型间流转的安全合规机制是否完善?
Lire le sujet complet →
Continuer la recherche →