微軟最新AI模型Fara-7B,強調本地端運行與資料隱私,企業可安心使用。Fara-7B具備卓越效能,透過視覺感知操作,完成複雜任務,展現AI模型的無限潛力。微軟於 11 月 24 日正式發表最新 AI 模型 Fara-7B,這款具備 70 億參數的模型被定位為「電腦使用代理」(Computer Use Agent, CUA),主打可直接在本地端運行、無需依賴雲端算力,同時兼顧高效能與資料隱私。 主打企業資料安全,支援「視覺感知」操作 Fara-7B 的設計核心在於滿足企業客戶對敏感資訊處理的隱私與合規需求。由於模型足夠精簡,可以在個人電腦上執行,不僅降低延遲,也避免資料上傳至雲端,有助於落實本地化自動化,例如用於內部帳號管理、機密文件處理等場景。 這款模型最大特色是採用「看螢幕操作」方式與網站互動——透過螢幕截圖讀取畫面排版,再預測滑鼠點擊、輸入或滾動等動作。與傳統依賴瀏覽器結構的方式不同,Fara-7B 完全基於像素級資料進行推理,因此即便是程式碼結構混亂的網站,也能正常運作。 微軟研究院產品經理 Yash Lara 表示,Fara-7B 透過本地端處理視覺輸入,實現所謂的「像素主權」,讓包括醫療、金融等高度監管產業也能安心使用。 實測效能超越 GPT-4o,小模型更高效 在 WebVoyager 測試基準中,Fara-7B 的任務完成率達 73.5%,高於 GPT-4o 的 65.1% 與 UI-TARS-1.5-7B 的 66.4%。此外,Fara-7B 完成任務平均僅需 16 步操作,明顯優於 UI-TARS-1.5-7B 的 41 步,在準確率與效率之間達成最佳平衡。 Fara-7B 同時引入「關鍵確認點」(critical checkpoints)機制,在遇到涉及用戶個資或不可逆操作(如發送信件、金錢轉移)時會自動暫停並請求確認,搭配「Magentic-UI」互動介面,提供人機協作的安全防線。 知識蒸餾與專家示範訓練,強化自主學習潛力 Fara-7B 採用「知識蒸餾」訓練方法,整合由多代理系統 Magentic-One 所產生的 14.5 萬筆成功導航範例,並壓縮至單一模型中學習。此外,底層模型基於 Qwen2.5-VL-7B,擁有最長 128,000 token 的上下文窗口,具備優異的圖文對齊能力,訓練過程以模仿人類專家操作為主。 微軟表示,未來不會盲目追求更大模型,而是致力於打造更「小而聰明、安全」的模型,並計畫引入強化學習(RL)於合成沙箱環境中進行自學訓練。 已開源上架,可自由測試商用但尚非正式產品 目前 Fara-7B 已透過 MIT 授權開源釋出,可在 Hugging Face 與微軟 Foundry 平台下載使用,允許用於商業應用。但微軟也提醒,該模型尚未達到生產環境部署標準,目前主要適合開發者用於原型測試與功能驗證。   延伸閱讀:Google 推 WeatherNext 2 新一代氣象預報 AI 模型,Pixel、搜尋、Gemini 搶先用 延伸閱讀:研究者發現讓 AI 變得更有創意的懶人提問法,不論 ChatGPT、Gemini 等任何 AI 模型皆能適用 延伸閱讀:Anthropic 發表Claude Haiku 4.5 小型 AI 模型:僅 1/3 成本、效能對標 Sonnet 4,程式表現甚至小贏  加入T客邦Facebook粉絲團微軟最新AI模型Fara-7B,強調本地端運行與資料隱私,企業可安心使用。Fara-7B具備卓越效能,透過視覺感知操作,完成複雜任務,展現AI模型的無限潛力。微軟於 11 月 24 日正式發表最新 AI 模型 Fara-7B,這款具備 70 億參數的模型被定位為「電腦使用代理」(Computer Use Agent, CUA),主打可直接在本地端運行、無需依賴雲端算力,同時兼顧高效能與資料隱私。 主打企業資料安全,支援「視覺感知」操作 Fara-7B 的設計核心在於滿足企業客戶對敏感資訊處理的隱私與合規需求。由於模型足夠精簡,可以在個人電腦上執行,不僅降低延遲,也避免資料上傳至雲端,有助於落實本地化自動化,例如用於內部帳號管理、機密文件處理等場景。 這款模型最大特色是採用「看螢幕操作」方式與網站互動——透過螢幕截圖讀取畫面排版,再預測滑鼠點擊、輸入或滾動等動作。與傳統依賴瀏覽器結構的方式不同,Fara-7B 完全基於像素級資料進行推理,因此即便是程式碼結構混亂的網站,也能正常運作。 微軟研究院產品經理 Yash Lara 表示,Fara-7B 透過本地端處理視覺輸入,實現所謂的「像素主權」,讓包括醫療、金融等高度監管產業也能安心使用。 實測效能超越 GPT-4o,小模型更高效 在 WebVoyager 測試基準中,Fara-7B 的任務完成率達 73.5%,高於 GPT-4o 的 65.1% 與 UI-TARS-1.5-7B 的 66.4%。此外,Fara-7B 完成任務平均僅需 16 步操作,明顯優於 UI-TARS-1.5-7B 的 41 步,在準確率與效率之間達成最佳平衡。 Fara-7B 同時引入「關鍵確認點」(critical checkpoints)機制,在遇到涉及用戶個資或不可逆操作(如發送信件、金錢轉移)時會自動暫停並請求確認,搭配「Magentic-UI」互動介面,提供人機協作的安全防線。 知識蒸餾與專家示範訓練,強化自主學習潛力 Fara-7B 採用「知識蒸餾」訓練方法,整合由多代理系統 Magentic-One 所產生的 14.5 萬筆成功導航範例,並壓縮至單一模型中學習。此外,底層模型基於 Qwen2.5-VL-7B,擁有最長 128,000 token 的上下文窗口,具備優異的圖文對齊能力,訓練過程以模仿人類專家操作為主。 微軟表示,未來不會盲目追求更大模型,而是致力於打造更「小而聰明、安全」的模型,並計畫引入強化學習(RL)於合成沙箱環境中進行自學訓練。 已開源上架,可自由測試商用但尚非正式產品 目前 Fara-7B 已透過 MIT 授權開源釋出,可在 Hugging Face 與微軟 Foundry 平台下載使用,允許用於商業應用。但微軟也提醒,該模型尚未達到生產環境部署標準,目前主要適合開發者用於原型測試與功能驗證。   延伸閱讀:Google 推 WeatherNext 2 新一代氣象預報 AI 模型,Pixel、搜尋、Gemini 搶先用 延伸閱讀:研究者發現讓 AI 變得更有創意的懶人提問法,不論 ChatGPT、Gemini 等任何 AI 模型皆能適用 延伸閱讀:Anthropic 發表Claude Haiku 4.5 小型 AI 模型:僅 1/3 成本、效能對標 Sonnet 4,程式表現甚至小贏  加入T客邦Facebook粉絲團

微軟推出 Fara-7B 小型 AI 模型,在地端直接執行、效能超越 GPT-4o

2025/11/28 13:30

微軟於 11 月 24 日正式發表最新 AI 模型 Fara-7B,這款具備 70 億參數的模型被定位為「電腦使用代理」(Computer Use Agent, CUA),主打可直接在本地端運行、無需依賴雲端算力,同時兼顧高效能與資料隱私。

主打企業資料安全,支援「視覺感知」操作

Fara-7B 的設計核心在於滿足企業客戶對敏感資訊處理的隱私與合規需求。由於模型足夠精簡,可以在個人電腦上執行,不僅降低延遲,也避免資料上傳至雲端,有助於落實本地化自動化,例如用於內部帳號管理、機密文件處理等場景。

這款模型最大特色是採用「看螢幕操作」方式與網站互動——透過螢幕截圖讀取畫面排版,再預測滑鼠點擊、輸入或滾動等動作。與傳統依賴瀏覽器結構的方式不同,Fara-7B 完全基於像素級資料進行推理,因此即便是程式碼結構混亂的網站,也能正常運作。

微軟研究院產品經理 Yash Lara 表示,Fara-7B 透過本地端處理視覺輸入,實現所謂的「像素主權」,讓包括醫療、金融等高度監管產業也能安心使用。

實測效能超越 GPT-4o,小模型更高效

在 WebVoyager 測試基準中,Fara-7B 的任務完成率達 73.5%,高於 GPT-4o 的 65.1% 與 UI-TARS-1.5-7B 的 66.4%。此外,Fara-7B 完成任務平均僅需 16 步操作,明顯優於 UI-TARS-1.5-7B 的 41 步,在準確率與效率之間達成最佳平衡。

Fara-7B 同時引入「關鍵確認點」(critical checkpoints)機制,在遇到涉及用戶個資或不可逆操作(如發送信件、金錢轉移)時會自動暫停並請求確認,搭配「Magentic-UI」互動介面,提供人機協作的安全防線。

知識蒸餾與專家示範訓練,強化自主學習潛力

Fara-7B 採用「知識蒸餾」訓練方法,整合由多代理系統 Magentic-One 所產生的 14.5 萬筆成功導航範例,並壓縮至單一模型中學習。此外,底層模型基於 Qwen2.5-VL-7B,擁有最長 128,000 token 的上下文窗口,具備優異的圖文對齊能力,訓練過程以模仿人類專家操作為主。

微軟表示,未來不會盲目追求更大模型,而是致力於打造更「小而聰明、安全」的模型,並計畫引入強化學習(RL)於合成沙箱環境中進行自學訓練。

已開源上架,可自由測試商用但尚非正式產品

目前 Fara-7B 已透過 MIT 授權開源釋出,可在 Hugging Face 與微軟 Foundry 平台下載使用,允許用於商業應用。但微軟也提醒,該模型尚未達到生產環境部署標準,目前主要適合開發者用於原型測試與功能驗證。

  • 延伸閱讀:Google 推 WeatherNext 2 新一代氣象預報 AI 模型,Pixel、搜尋、Gemini 搶先用
  • 延伸閱讀:研究者發現讓 AI 變得更有創意的懶人提問法,不論 ChatGPT、Gemini 等任何 AI 模型皆能適用
  • 延伸閱讀:Anthropic 發表Claude Haiku 4.5 小型 AI 模型:僅 1/3 成本、效能對標 Sonnet 4,程式表現甚至小贏
Disclaimer: The articles reposted on this site are sourced from public platforms and are provided for informational purposes only. They do not necessarily reflect the views of MEXC. All rights remain with the original authors. If you believe any content infringes on third-party rights, please contact service@support.mexc.com for removal. MEXC makes no guarantees regarding the accuracy, completeness, or timeliness of the content and is not responsible for any actions taken based on the information provided. The content does not constitute financial, legal, or other professional advice, nor should it be considered a recommendation or endorsement by MEXC.

You May Also Like

Grayscale debuts first US spot crypto ETPs with staking

Grayscale debuts first US spot crypto ETPs with staking

The post Grayscale debuts first US spot crypto ETPs with staking appeared on BitcoinEthereumNews.com. Grayscale Investments has just launched the first US-listed spot crypto exchange-traded products (ETPs) offering staking. The Grayscale Ethereum Trust ETF (ETHE) and Grayscale Ethereum Mini Trust ETF (ETH) now enable Ether staking, while the Grayscale Solana Trust (GSOL) has activated staking capabilities ahead of its proposed uplisting as a spot Solana ETP. The move provides traditional brokerage investors with access to staking rewards — previously confined to native crypto platforms — through regulated vehicles. The products are not registered under the Investment Company Act of 1940, meaning they operate outside the framework governing traditional mutual funds and ETFs. Staking, the process of locking up tokens to secure proof-of-stake blockchains like Ethereum and Solana in exchange for rewards, introduces yield potential but also adds operational and network risks.  Grayscale said staking will be managed through institutional custodians and diversified validator networks to reduce single-party risk. This marks the first time US investors can access staking yield through exchange-traded exposure to Ethereum and Solana, expanding upon regulatory acceptance that began with spot Bitcoin ETFs in January 2024 and spot Ether ETFs in July 2024.  Grayscale CEO Peter Mintzberg called the initiative “first mover innovation,” underscoring the firm’s role in shaping institutional crypto access. This is a developing story. This article was generated with the assistance of AI and reviewed by editor Jeffrey Albus before publication. Get the news in your inbox. Explore Blockworks newsletters: Source: https://blockworks.co/news/grayscale-us-spot-crypto-etps-staking
Share
BitcoinEthereumNews2025/10/06 21:29
Spot XRP ETFs Nears $1B AUM Milestone as Streak of No Outflows Continues

Spot XRP ETFs Nears $1B AUM Milestone as Streak of No Outflows Continues

The post Spot XRP ETFs Nears $1B AUM Milestone as Streak of No Outflows Continues appeared on BitcoinEthereumNews.com. The U.S. Spot XRP ETFs is now near the $1 billion mark of assets under management in less than a month since their launch. This follows from the product maintaining consistent inflows with no single outflow recorded yet. XRP ETFs See Continuous Inflows Since Launch Since its first launch on November 14, spot XRP funds have seen continued inflows. According to data from SoSoValue, the total inflows into these funds have now risen to $881.25 million. The funds attracted $12.84 million of new money yesterday. The daily trading volumes remained stable at $26.74 million. Source: SoSoValue Reaching nearly $1 billion in less than 30 days makes the product among the fastest growing crypto investment products in the United States. Notably, Spot Solana ETFs also accumulated over $600 million since their launch. On the other hand, Bitcoin and Ethereum ETFs are holding about $58 billion and about $13 billion in assets under management respectively. Much of the early growth traces back to the first Canary Capital’s XRP ETF. Its opening on November 13 brought one of the strongest crypto ETF openings to date. It saw more than $59 million in first-day trading volume and $245 million in net inflows. Shortly after Canary’s launch, firms like Grayscale, Bitwise, and Franklin Templeton introduced their own XRP products. Bitwise’s fund also did well on its launch, recording over $105 million in early inflows. Meanwhile, the market is getting ready for yet another addition. 21Shares’ U.S. spot XRP fund also got the green light from the SEC. It will trade under the ticker TOXR on the Cboe BZX Exchange. XRP Products Keep Gaining Momentum in the Market The token’s funds continued to expand this week. REX Shares and Tuttle Capital have launched the T-REX 2X Long XRP Daily Target ETF. This new ETF allows traders…
Share
BitcoinEthereumNews2025/12/05 14:11
Headwind Helps Best Wallet Token

Headwind Helps Best Wallet Token

The post Headwind Helps Best Wallet Token appeared on BitcoinEthereumNews.com. Google has announced the launch of a new open-source protocol called Agent Payments Protocol (AP2) in partnership with Coinbase, the Ethereum Foundation, and 60 other organizations. This allows AI agents to make payments on behalf of users using various methods such as real-time bank transfers, credit and debit cards, and, most importantly, stablecoins. Let’s explore in detail what this could mean for the broader cryptocurrency markets, and also highlight a presale crypto (Best Wallet Token) that could explode as a result of this development. Google’s Push for Stablecoins Agent Payments Protocol (AP2) uses digital contracts known as ‘Intent Mandates’ and ‘Verifiable Credentials’ to ensure that AI agents undertake only those payments authorized by the user. Mandates, by the way, are cryptographically signed, tamper-proof digital contracts that act as verifiable proof of a user’s instruction. For example, let’s say you instruct an AI agent to never spend more than $200 in a single transaction. This instruction is written into an Intent Mandate, which serves as a digital contract. Now, whenever the AI agent tries to make a payment, it must present this mandate as proof of authorization, which will then be verified via the AP2 protocol. Alongside this, Google has also launched the A2A x402 extension to accelerate support for the Web3 ecosystem. This production-ready solution enables agent-based crypto payments and will help reshape the growth of cryptocurrency integration within the AP2 protocol. Google’s inclusion of stablecoins in AP2 is a massive vote of confidence in dollar-pegged cryptocurrencies and a huge step toward making them a mainstream payment option. This widens stablecoin usage beyond trading and speculation, positioning them at the center of the consumption economy. The recent enactment of the GENIUS Act in the U.S. gives stablecoins more structure and legal support. Imagine paying for things like data crawls, per-task…
Share
BitcoinEthereumNews2025/09/18 01:27