aifollow.news
返回 IT之家 科技新闻
IT之家 科技新闻· · 原发布时间 AI 评分60

微软发布 MAI Code 2 与 Flash,并在 GitHub Copilot 上线

AI 辅助摘要

微软推出两款自研编程模型:MAI Code 2 面向复杂开发,Flash 侧重解释代码、定点修改等高频任务。Code 2 向符合条件的 Copilot Pro、Business 和 Enterprise 订阅用户开放;Flash 还覆盖 Free 和 Student 用户。微软公布的四项编程基准平均结果显示,Code 2 表现约为 GPT-6.1 Sol 的 98%,价格约为其 40%。

正文 · AI 翻译

Thanks to IT Home reader 华南吴彦祖 for the tip!

IT Home reported on November 18 that Microsoft unveiled its in-house coding models MAI Code 2 and MAI Code 2 Flash at today's Ignite 2026 event and made them available on GitHub Copilot.

IT Home has included the relevant video below:

The 2 products target complex software development and frequent everyday coding tasks, respectively, with a focus on striking a better balance between code quality and cost.

Code 2 is positioned as a general-purpose coding model that can be used to investigate bugs in depth, modify code across multiple files, understand an entire code repository, and develop user interfaces. Microsoft says the model emphasizes providing practical code changes directly, reducing unnecessary back-and-forth between developers and AI and helping get code into deployment.

Code 2 Flash focuses more on lightweight, frequent tasks, such as explaining code, making targeted changes, fixing common problems, and iterating quickly. Both models support workflows in which AI calls tools and advances development tasks step by step.

Value for money is a main selling point of this release. Microsoft says that, averaged across the 4 coding benchmarks it published, Code 2 delivers about 98% of GPT-6.1 Sol's performance at about 40% of its price.

Official test data shows that MAI Code 2 costs just $0.309 per task on average, substantially less than GPT-6.1 Sol's $0.761 and only about 5% of Claude Opus 5.5's $6.306. While maintaining top-tier engineering performance, it helps businesses substantially reduce their AI compute bills.

With a 50% promotional discount, the lightweight and efficient MAI Code 2 Flash costs as little as $0.052 per task on average. Its performance matches or even surpasses competing flagship lightweight models, while its cost is just 65% of GPT-6 Luna's and 20% of Claude Haiku 5.5's.

Across the 3 major benchmark tests DeepSWE 1.1, Terminal-Bench 2.1, and SWE Atlas, MAI Code 2 Flash achieved an average pass rate of 61.0%, surpassing GPT-6 Luna (55.1%) and Claude Haiku 5.5 (53.9%). Delivering higher-quality code at a lower price, it redefines the benchmark for value among lightweight coding models.

In terms of availability, Code 2 is offered to eligible subscribers, including Copilot Pro, Business, and Enterprise users; Flash is also available to Free and Student users. Enterprise developers can choose between the models based on task difficulty and usage volume.

发现内容有误?提交纠错