该公司目前正在积极推动开放权重模型。在一封题为《开放权重与美国 AI 领导力》的公开信中（https://www.microsoft.com/en-us/corporate-responsibility/topics/open-weight/），微软与 Meta、Nvidia、Hugging Face、Mistral 及其他20多家公司共同签署了该信，信中指出，“领导力的评判标准不在于某个前沿 AI 模型，而在于美国是否建立了一个强大、开放的生态系统，并将其扩展到各个行业。”
开放权重模型旨在激发创新并减少对少数供应商的依赖。该信还将“蒸馏”——即较小的 AI 模型从更强大的模型输出中学习——定义为构建 AI 的合法方式（https://the-decoder.com/nadella-calls-out-ai-labs-like-openai-and-anthropic-for-banning-distillation-while-training-on-everyone-elses-data/）。信中写道：“这反映了从现有技术中学习、在其基础上构建并改进的悠久传统，这一传统自开源软件运动兴起以来一直推动着创新。”
微软甚至没有尝试掩饰这是为了降低成本。公司在自己的博客中指出：https://microsoft.ai/news/hill-climbing-mai-models-for-github-copilot-and-excel/，较小的 MAI 模型可以在较旧的 Nvidia GPU（如 H100 和 A100）上运行，而不需要最新的加速器，这“显著降低了微软的部署成本”。未来还计划在 Copilot 聊天、Outlook 和 PowerPoint 上推出更多 MAI 模型。广告
保持人工智能领域的最新动态。清晰、有用、无冗余。
关注 The Decoder 获取人工智能新闻、背景故事和专家分析。
The Decoder：https://the-decoder.com/
Microsoft has figured out it's better positioned as an AI orchestrator than as an OpenAI reseller.
The company is now actively pushing for open-weight models. In an open letter：https://www.microsoft.com/en-us/corporate-responsibility/topics/open-weight/ titled "Open Weights and American AI Leadership," which Microsoft signed alongside Meta, Nvidia, Hugging Face, Mistral, and more than 20 other companies, the group argues that "leadership will be judged not by one frontier AI model, but by whether the United States builds a strong, open ecosystem that diffuses into every sector."
Open-weight models are supposed to spur innovation and reduce dependence on a handful of providers. The letter also frames "distillation," in which a smaller AI model learns from a more capable model's output, as a legitimate way to build AI：https://the-decoder.com/nadella-calls-out-ai-labs-like-openai-and-anthropic-for-banning-distillation-while-training-on-everyone-elses-data/. "It reflects a long tradition of learning from, building upon, and improving existing technologies, a tradition that has helped drive innovation since the rise of the open-source software movement," the letter states. Ad
Chinese providers have faced particular criticism for using distillation：https://the-decoder.com/kimi-k3-trails-frontier-us-models-by-a-wide-margin-on-cyber-exploits-and-distillation-may-explain-why/, while the Trump administration reportedly plans to take action against Chinese open-weight models：https://the-decoder.com/trump-administration-reportedly-builds-a-slow-motion-ban-on-chinese-ai-models-through-sanctions-and-soft-pressure/. That's likely the main reason Microsoft and the other signatories are speaking up now. Ad DEC_D_Incontent-1
Microsoft's enthusiasm for openness makes more sense when you look at its business model. The company makes money from cloud infrastructure on Azure and the lock-in created by its Windows and Office ecosystem. More models on Azure give customers fewer reasons to use rival AI platforms. Relying less on expensive models from OpenAI and Anthropic also improves Microsoft's margins.
A fragmented model market also stops any single AI lab from becoming powerful enough to threaten Microsoft's cloud and operating system businesses. CEO Satya Nadella wrote on LinkedIn：https://www.linkedin.com/posts/satyanadella_in-a-world-where-software-has-real-marginal-share-7486096175440347137-yDDh/?utm_source=share&utm_medium=member_desktop&rcm=ACoAABfX0nABz6sCWbPldiV_9liETVfz5fRLAD0 that the key is using "the right model for each task" while keeping costs down. He had previously warned that a few AI models could absorb the value of entire industries：https://the-decoder.com/microsoft-ceo-satya-nadella-warns-of-a-small-number-of-ai-systems-capturing-all-the-economic-returns/. Ad
Microsoft is also replacing OpenAI and Anthropic models in GitHub Copilot, Excel, and Outlook with its in-house MAI family. Whether customers benefit from that swap is questionable：https://the-decoder.com/copilot-goes-cheap-as-microsoft-phases-out-openai-and-anthropic-models-to-cut-costs/. Independent benchmarks put MAI well behind OpenAI and Anthropic, with performance roughly matching Deepseek V3.2.
Copilot customers are getting weaker AI for the same price, while Microsoft pockets better margins. The company claims MAI performs just as well or better, but only compares it against smaller, less powerful models like GPT-5.4 Mini, Anthropic's Haiku, and "GPT-5.6" without specifying whether that means Sol, Terra, Luna, or which reasoning mode it runs at. That's not how you do benchmarks, and Microsoft's "superintelligence team," which made the comparison, certainly knows it. Ad DEC_D_Incontent-2
Microsoft doesn't even try to hide that this is about cutting costs. The company states on its own blog：https://microsoft.ai/news/hill-climbing-mai-models-for-github-copilot-and-excel/ that the smaller MAI model runs on older Nvidia GPUs like the H100 and A100 instead of requiring the latest accelerators, which "significantly lowers the cost of deployment for Microsoft." More MAI models are planned for Copilot Chat, Outlook, and PowerPoint. Ad
Stay in the loop on AI. Clear, useful, no fluff.
Follow The Decoder for AI news, background stories and expert analyses.
扫码直达文章详情