跳到主要內容

Copilot多模型評測:微軟砸20億美元導入Claude | Copilot Multi-Model Review: Microsoft Pays Anthropic $2B

By Kit 小克 | AI Tool Observer | 2026-10-13

🇹🇼 Copilot多模型評測:微軟砸20億美元導入Claude

Copilot 這次真的變了。微軟不再只靠 OpenAI 一家撐場,而是傳出每年砸下至少 20億美元 授權費,把 Anthropic 的 Claude 正式請進 Word、Excel、PowerPoint、Outlook,甚至 Copilot Cowork 跟 Copilot Studio,讓用戶能在同一個介面裡自由切換要用 GPT 還是 Claude 做事。這波「多模型 Copilot」策略,對一般用戶跟企業 IT 來說,到底是加分還是多一層麻煩?

Copilot 多模型是什麼:不是取代,是加選項

Claude 的 Office 外掛其實從今年 5 月就已經在 Word、Excel、PowerPoint 正式上線(GA),Outlook 整合當時還在公開測試。9 月之後,微軟把 Claude Opus 5.5 跟 OpenAI 的 GPT-6 Sol 一起排進 Copilot 的模型選單,涵蓋 Chat、Cowork、Copilot Studio。換句話說,Copilot 從「單一引擎」變成「可換引擎」的工具,使用者可以依任務挑模型,而不是被綁死在一家供應商身上。

微軟付的 20 億美元買了什麼

這筆錢不是單純的授權費而已,背後是更大的交易結構:Anthropic 承諾在 Azure 上採購 300 億美元運算資源,微軟反過來投資 Anthropic 最高 50 億美元,Nvidia 再加碼 100 億美元。三方互相綁定,微軟等於用真金白銀買下一個「不只靠 OpenAI」的保險。

實際用起來差在哪

  • Excel 財務函數自動化:官方跟多家評測都指出 Claude 在這類需要多步驟推理的任務表現較穩
  • PowerPoint 依指令生成整份簡報:Claude 版本在排版邏輯跟內容結構上出錯率較低
  • 一般寫信、整理會議紀要:兩家模型差異不大,選哪個都可以

IT 部門該注意的風險

多一個模型供應商,代表企業要多盤點一套資料治理規則。Directions on Microsoft 的報告直接點名,這是「多了選擇、也多了風險」——兩家公司的資料保留政策、安全認證、地區合規都不盡相同,IT 管理員要搞懂兩套規則,而不是一套。某些租戶甚至預設就把 Anthropic 模型開啟了,等於還沒評估風險就已經在用。

老實說,這則新聞的重點不是「哪個模型更強」,而是微軟親手打破了自己對 OpenAI 的獨家依賴。這是一場避險操作,不是技術突破,用戶拿到的是「選擇權」,但相對也要自己花時間比較哪個模型適合哪個任務。Claude 跟 Copilot 綁在一起的消息聽起來很熱鬧,但實際上要不要切換、哪個場景該用哪個模型,還是得自己動手測。

好不好用,試了才知道。


🇺🇸 Copilot Multi-Model Review: Microsoft Pays Anthropic $2B

Copilot just stopped being an OpenAI-only product. Microsoft is reportedly paying Anthropic at least $2 billion a year to bring Claude into Word, Excel, PowerPoint, Outlook, Copilot Cowork, and Copilot Studio, letting users pick GPT or Claude from the same interface. Is this multi-model Copilot shift actually useful, or just another layer of complexity for IT?

What Multi-Model Copilot Actually Means

Claude add-ins for Word, Excel, and PowerPoint went GA back in May, with Outlook in public beta at the time. By September, Microsoft slotted Claude Opus 5.5 alongside OpenAI's GPT-6 Sol into the Copilot model picker across Chat, Cowork, and Copilot Studio. Copilot went from a single-engine tool to a swappable one: pick the model per task instead of being locked to one vendor.

What the $2B a Year Actually Buys

This isn't just a licensing fee, it sits inside a bigger deal: Anthropic committed to $30 billion in Azure compute purchases, Microsoft invested up to $5 billion back into Anthropic, and Nvidia added another $10 billion. Microsoft is effectively buying itself insurance against over-relying on OpenAI alone.

Where the Models Actually Differ

  • Excel financial automation: Multiple reviews note Claude handles multi-step reasoning tasks more reliably
  • PowerPoint generation from instructions: Claude's output tends to have fewer layout and structure errors
  • Everyday email drafting and meeting notes: barely any difference, pick either

The Risk IT Teams Need to Watch

Adding a second model vendor means a second set of data-governance rules to audit. A report from Directions on Microsoft bluntly frames this as choice and risk at the same time: the two companies differ on data retention, security certifications, and regional compliance, so admins now have to understand two rulebooks instead of one. Some tenants reportedly had Anthropic models enabled by default before anyone ran a risk review.

Honestly, the real story here isn't which model wins, it's Microsoft deliberately breaking its own exclusive dependence on OpenAI. This is a hedge, not a breakthrough. Users get optionality, but that means actually testing which model fits which task instead of assuming either one is simply better. The Claude-in-Copilot headline sounds exciting, but whether you should switch, and when, still comes down to hands-on testing.

好不好用,試了才知道。

Sources / 資料來源

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

Google Ironwood TPU v7 推理專用晶片解析:效能追平 NVIDIA、成本低 44%,AI 晶片戰爭正式開打 | Google Ironwood TPU v7 Explained: Matching NVIDIA Performance at 44% Lower Cost — The AI Chip War Heats Up

Claude Code 實測:AI 幫你寫程式到底行不行? | Claude Code Review: Can AI Really Code for You?

Cursor vs GitHub Copilot vs Claude Code:AI 程式助手大比拼 | AI Coding Assistants Compared: Cursor vs GitHub Copilot vs Claude Code