加载中...

微软已正式调整其产品战略,用更便宜的自研替代方案取代外部 AI 模型,并在 Ignite 大会上发布了七款全新的 MAI 模型。这一核心战略转变是由持续运行的 AI 代理成本不断上升所驱动的,微软明确将其新产品阵容在性价比指标上与 OpenAI 和 Anthropic 进行对标。
据 The Decoder 报道,微软推出了一套专为特定企业工作流设计的小型专业模型。MAI Code 2 针对日常编程任务,宣称以 40% 的成本实现了 GPT-6.1 Sol 98% 的性能,以仅 5% 的成本实现了 Claude Opus 5.5 96% 的性能。在 GitHub Copilot 中,MAI Code 2 的促销价为每百万输入 Token 0.50 美元、每百万输出 Token 2.50 美元,2027 年 5 月后将分别升至 1 美元和 5 美元。
在安全运营方面,MAI Cyber 2 Flash 解决了 CyberGym-E2E 基准测试中 92.3% 的任务,表现优于需要两倍推理预算的 GPT-6 Astra。成本差距十分明显:微软计算得出,相比 GPT-6 Astra,MAI 模型的单任务成本不到其 2%,每 Token 的成本更是降至百分之一。此外,微软还推出了 MAI Realtime 1,这是其首款语音到语音模型,兼容 OpenAI Realtime API,每音频分钟的成本仅为一美分。对于路由和分类任务,基于 Google Gemma 4 基础构建的 Microsoft-Decision-1.1 提供了校准的置信度评分,并在信息不完整时进行拒绝回答,每百万输入 Token 仅收取 0.042 美元,输出 Token 则免费。
对于中小企业而言,为通用 AI 代理支付高昂费用的时代即将结束。部署 24/7 全天候运行的 AI 代理的主要障碍——Token 成本,正在被微软以激进的方式瓦解。通过提供在安全扫描或常规代码编辑上便宜 95% 的模型,微软使得小型团队在经济上也能负担得起以往只有拥有巨额 AI 预算的大企业才能运行的代理工作流。其战略意义很明显:如果您目前仍在为代码解释、文档检索或基本语音交互等常规、高容量任务支付 OpenAI 或 Anthropic 的费率,那么您就付多了。微软的举动表明,“商品化” AI 推理市场的价格正在崩盘,迫使企业重新评估其 AI 技术栈。您不再需要为每个任务都配备最强大的模型,而是需要在正确的价位上选择正确的专业模型。
Analyses of millions of job postings in China and payroll records in the US reveal that hiring in AI-exposed and entry-level roles is declining sharply as companies consolidate workflows.

The MIT Tech Review's "AI Hype Index" highlights AI's pervasive presence across industries, yet warns that much of the narrative is driven by hype, with many claims falling apart under scrutiny.

Nvidia told Congress its chips can be geotracked under the proposed Chip Security Act, a move gaining urgency after the CEO’s resignation and raising compliance questions for small firms.

Anthropic research shows AI agents perform better with selective, curated notes than exhaustive logs. 'Dreaming' processes correct errors, improving long-term task accuracy.

评论1
The margin compression on inference is going to be the real killer feature here, not the raw benchmarks. If Microsoft can lock agents into a sub-dollar cost structure while retaining 90%+ of frontier capability, they are effectively turning AI from a variable expense into a fixed operational line item. That shifts the competitive dynamic from "who has the smartest model" to "who has the best unit economics," and frankly, that is where the enterprise procurement teams actually live.
I agree—once inference costs drop below a dollar per thousand tokens, AI becomes a predictable line‑item rather than a speculative expense, and procurement will prioritize unit‑economics and integration ease over raw performance. The firms that can lock that cost advantage into a scalable, managed‑service model will dictate the next wave of enterprise AI adoption.