Microsoft AI Releases Microsoft-Decision-1: A Qwen3.5-9B Decision-Scoring Model
https://www.marktechpost.com/2026/10/09/microsoft-ai-releases-microsoft-decision-1-a-qwen3-5-9b-decision-scoring-model/📌 【微軟跟進新賽道】Microsoft-Decision-1：Qwen3.5-9B 打造的決策評分模型
TL;DR：微軟推出「決策模型」新類別產品，標榜低延遲與高準確率，但競品 benchmark 方法論已被對手公開質疑。
就在開源陣營推出決策模型沒多久，微軟也端出自己的版本，而且順手把對手的測試方法拿出來比較，意外掀起一場 benchmark 公平性的爭論。
🤔 什麼是決策模型
Microsoft-Decision-1 讀取一個情境、一個問題與固定的答案選項，回傳每個選項的校正機率，而不是生成文字。微軟稱這是一個全新的 AI 類別，目標是產出「軟體可以直接拿來用」的輸出，而不是需要人類再解讀的自然語言。這個模型是以阿里巴巴的 Qwen3.5-9B 為基礎做 post-training 得到的，目前已在 Microsoft Foundry 與 OpenRouter 上架。訓練資料來自微軟 Open Data 流程下的公開資料集，加上合成資料，輸出格式固定為 JSON。微軟表示未來版本計畫改以 MAI 與 OpenAI 的模型作為基底。OpenRouter 上的說明指出，權重會持續更新，但 API 介面會保持不變。
📊 36 項 benchmark、14 萬題測試，延遲是強打重點
微軟在 36 項 benchmark、共 147,137 道題目上比較了 9 套系統，且這些 benchmark 在訓練階段是完全隔絕的。Microsoft-Decision-1 的平均準確率以 83.5% 排名第一，p50 延遲為 85 毫秒、p95 為 125 毫秒，比 Quyet-1.0-Large 快 4.5 倍，比 GPT-6 Sol（3.01 秒）快 35 倍。微軟也做了穩健性測試：對每個請求做 8 種擾動（包含改寫句子、選項洗牌），結果平均只有 1.3% 的擾動會讓模型翻轉決策；當擾動方式是改寫、反向排列或洗牌選項時，翻轉率甚至是零。在安全性測試上，微軟針對 11 項涵蓋有害內容、jailbreak 與 prompt injection 的 benchmark 跑了 5,250 次請求，宣稱能正確拒絕危險請求同時保留大部分可用性，但沒有公開具體分數。
💡 延遲比較方法被對手公開打臉
這組漂亮的延遲數字有個但書：微軟是透過自家 Foundry 直接量測自己的模型，但圖表上競品的延遲數字，用的是 JevBench v1.6.1 的「調整後中位數」，不是原始量測值。H2O.ai 在自家 model card 上直接點名反駁，指出 JevBession 的調整方式會把實際耗時加倍，再外加 0.15 秒，並宣稱自家模型實際測得的中位數是 29 毫秒，不是圖表上顯示的 210 毫秒。另外值得注意，Microsoft-Decision-1 目前並未出現在 JevBench 官方排行榜上；該榜單的官方綜合分數裡，H2O-Lightning-4B 排名第一，拿下 72.5 分。這代表讀者在看任何決策模型的「比較圖」時，最好先確認延遲與分數是不是用同一套量測方法算出來的。
⚠️ 使用上的限制
開發者可以在 Microsoft Foundry 以「Direct from Azure」的正式服務呼叫這個模型；但在 OpenRouter 上，它跑在獨立的 Decisions API，而不是一般的 chat completions 端點，意味著既有的 chat SDK 無法直接沿用。計價方面，輸入每百萬 token 收費 0.042 美元，輸出免費；作為對比，OpenAI 的 Luna decisions 端點輸入每百萬 token 收費 0.10 美元。
🎯 實務啟示
如果你的系統需要高頻率、低延遲的分類／路由／驗證任務，Microsoft-Decision-1 的輸入定價與延遲數字確實有吸引力；但在下決定前，建議自己用相同的量測方式重跑一次 benchmark，而不是完全依賴官方對比圖表。
🔗 來源
- 標題：Microsoft AI Releases Microsoft-Decision-1: A Qwen3.5-9B Decision-Scoring Model
- 作者／機構：Asif Razzaq, MarkTechPost
- 連結：https://www.marktechpost.com/2026/10/09/microsoft-ai-releases-microsoft-decision-1-a-qwen3-5-9b-decision-scoring-model/
#DecisionModel #Microsoft #Qwen #MicrosoftFoundry #OpenRouter #LLM #Benchmark #AIInfrastructure #MachineLearning #AgentTools
原始資料 MarkTechPost · 收集於 2026-10-10
摘要原文
Microsoft has released Microsoft-Decision-1 , a decision model for routing, classification, verification and agent control. Microsoft-Decision-1 is a decision-scoring model that returns a calibrated probability for each fixed answer option instead of generated text. It is post-trained from Alibaba’s Qwen3.5-9B and available now in Microsoft Foundry and OpenRouter . . A decision model reads an input and scores a closed set of options. It does not write prose. Microsoft frames decision models as a new AI category , built for outputs software can act on immediately. Microsoft post-trained Qwen3.5-9B for single-pass decision scoring . Given a situation, a question and fixed options, it returns a probability per option in one call. It plans to rebase future versions on MAI and OpenAI models. The Foundry model card lists the supported formats: Training used public datasets under Microsoft’s Open Data process plus synthetic data. Output is JSON. OpenRouter notes that weights update continually while the API shape stays fixed. Microsoft compared 9 systems across 36 benchmarks with 147,137 questions . Benchmarks were kept blind from training. Microsoft-Decision-1 led on average accuracy at 83.5%. Its p50 latency was 85 ms, with p95 at 125 ms. That is 4.5 times quicker than Quyet-1.0-Large and 35 times quicker than GPT-6 Sol , which took 3.01 s. Microsoft also tested robustness. It perturbs each request 8 ways, including paraphrasing and option shuffling. The model flipped its decision on 1.3% of perturbations on average. Flips were zero when options were paraphrased, reversed or shuffled. On safety, Microsoft ran 5,250 requests across 11 benchmarks. These covered harmful content, jailbreaks and prompt injection. Microsoft reports correct refusals with high retained utility, without publishing a score. Accuracy, calibration and latency rows come from Microsoft’s chart . Competitor latencies there use the JevBench v1.6.1 adjusted median. Not fully. Microsoft measured its own model through Foundry. Competitor figures use JevBench’s adjusted median, not raw timings. H2O.ai’s model card disputes this. It states that JevBench’s adjusted figure doubles measured time and adds 0.15 s. H2O says its model’s measured median is 29 ms, not 210 ms. Microsoft-Decision-1 does not appear on the JevBench board. On that board’s official composite, H2O-Lightning-4B ranks #1 at 72.5 . Developers can call it from Microsoft Foundry as a generally available ‘Direct from Azure’ model. On OpenRouter , it runs on the Decisions API, not the chat endpoint. Chat completions SDKs will not work. Pricing is $0.042 per million input tokens, with free output . OpenAI’s Luna decisions endpoint charges $0.10 per million input tokens . Check out the official announcement , the Foundry model card and the OpenRouter listing . All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. The post Microsoft AI Releases Microsoft-Decision-1: A Qwen3.5-9B Decision-Scoring Model appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成