solidot新版网站常见问题，请点击这里查看。
消息
本文已被查看 770 次
微软在 Qwen 模型基础上推出决策模型 Microsoft-Decision-1
微软加入了 Jev 粉丝俱乐部，推出了自己的决策模型。Jev 针对特定问题对大模型进行调优，减少“幻觉”，适用于各种商业应用场景。微软在阿里巴巴的开放权重模型 Qwen3.5-9B 基础上推出了 Microsoft-Decision-1。软件巨人表示未来会用自研模型及 OpenAI 的模型取代 Qwen 模型。微软称，其模型在延迟测试中速度分别是 H2O-Lightning-4B 的 2.5 倍和 Jev 的 2.8 倍；在 36 项基准测试中准确率（83.5%）位居榜首；置信度得分（92.2%）排名第二（仅次于 Quyet-1.0-Large）；在文本分类任务中，其成本不到 OpenAI GPT-6 Sol 的 1/20。
https://www.theregister.com/ai-and-ml/2026/10/10/microsoft-leans-on-open-weight-model-from-chinese-ai-lab-to-challenge-jev/5302473
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Click here to see the most common questions on the new version of the website.
Message
This has been viewed 770 times
Microsoft roll-out decision model based on Qwen
Microsoft joined the Jev Fan Club and launched its own decision model. Jev fine-tuning LLM in response to specific issues to reduce the “fantasia” that applies to various commercial applications. Microsoft launched Microsoft-Decisión-1 on the basis of Alibaba open weight model Qwen3.5-9B. The software giant indicates that the Qwen model will be replaced in the future with self-study models and Open AI models. Microsoft stated that its model was 2.5 times faster than H2O-Lightning-4B and 2.8 times more than Jev, respectively, in the delayed tests; ranked first in the 36 benchmark tests (83.5 per cent); ranked second in the confidence score (92.2 per cent) (after Quyet-1.0-Large); and cost less than 1/20 for Open AI GPT-6 Sol in the text classification missions.
https://www.theregister.com/ai-and-ml/2026/10/10/microsoft-leans-on-open-weight-model-from-chinese-ai-lab-to-challenge-jev/5302473
Auto-fetched and translated by AI Atlas. All rights belong to the original author.