Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Microsoft Plans Aion Instruct to Replace Phi Silica on Windows

AI summary

Microsoft plans to replace Phi Silica with Aion Instruct as Windows’ on-device language model, according to iThome. The 2.3-billion-parameter model expands the input context limit from 3,500 to 32,000 tokens, with phased Insider deployment starting November 20 and broader rollout to eligible release devices planned for late January 2027.

In-depth insight

Aion Instruct offers a much larger input window while allowing most applications to retain existing Windows AI APIs, though Microsoft’s performance and resource-use claims await a full technical report.

移至主內容

文/李建興|2026-11-16 發表

微軟**公布**Windows裝置端語言模型更替計畫,將以Aion Instruct取代Phi Silica,預計11月20日起先向參與Windows Insider計畫的Copilot+ PC分階段推送,2027年1月下旬擴大至符合硬體條件的正式版Windows裝置。新模型參數量減少,卻能處理更長的輸入內容,並將擴大支援Nvidia RTX及AMD RDNA 4系列GPU。

Phi Silica是微軟為Windows開發的小型語言模型,讓應用程式透過Windows AI API,在裝置上執行文字生成、摘要及改寫等工作,而無須將處理內容傳送至雲端。Aion Instruct接替後,多數應用程式仍可繼續使用既有API,不必修改程式。

模型規格方面,Phi Silica有38億個參數,Aion Instruct縮減至23億個,輸入上下文長度上限則從3,500個詞元提高至32,000個。更大的輸入容量讓應用程式能一次提供更多文件或對話內容,減少因長度限制而切分資料的需求。微軟表示新模型可改善執行效能與裝置資源使用,完整技術報告將於後續公布。

#Microsoft#Aion Instruct#Phi Silica#Windows#Windows AI APIs#Nvidia RTX#AMD RDNA 4

Original · 3 min readRead the original
Group chat

No comments yet. Start the conversation.