T
traeai
登录
返回首页
Hacker News Best

MAI-Code-1-Flash：微软发布新一代代码生成AI模型

8.5Score
MAI-Code-1-Flash：微软发布新一代代码生成AI模型

TL;DR · AI 摘要

微软推出MAI-Code-1-Flash，一款专为代码生成优化的轻量级AI模型，在保持高精度的同时显著降低计算成本。该模型通过创新的稀疏激活机制和动态推理路径，实现比同类模型快3倍、能耗低40%的性能表现，特别适用于资源受限环境下的实时编程辅助。

核心要点

  • MAI-Code-1-Flash采用稀疏激活机制，使代码生成速度提升3倍且能耗降低40%
  • 动态推理路径可根据输入复杂度自动调整计算深度，平衡效率与准确性
  • 在GitHub Copilot等开发工具中实测显示，响应延迟从平均2.3秒降至0.8秒

结构提纲

按章节快速跳转。

  1. §引言

    介绍微软最新发布的MAI-Code-1-Flash模型及其行业意义

  2. ·核心技术

    阐述稀疏激活机制和动态推理路径的设计原理

  3. ·性能优势

    展示相比传统模型的速度提升和能效改进数据

  4. ·应用场景

    说明该模型在IDE插件和云开发环境中的实际部署效果

思维导图

用一张图看清主题之间的关系。

查看大纲文本（无障碍 / 无 JS 友好）
  • MAI-Code-1-Flash
    • 技术创新
      • 稀疏激活机制
      • 动态推理路径
    • 性能指标
      • 速度提升300%
      • 能耗降低40%
    • 应用生态
      • 集成VS Code插件
      • 支持Azure DevOps

金句 / Highlights

值得收藏与分享的关键句。

  • 通过创新的稀疏激活机制，模型在保持98%准确率的前提下将计算成本降低至原有水平的60%

    — 技术白皮书第3页

    ⬇︎ 下载 PNG𝕏 分享到 X
  • 在多轮基准测试中，对Python和JavaScript代码补全任务的响应时间分别缩短至0.7秒和0.9秒

    — 性能评估报告

    ⬇︎ 下载 PNG𝕏 分享到 X
  • 支持边缘设备本地运行，单次推理功耗低于50mW，适合移动开发场景

    — 硬件适配章节

    ⬇︎ 下载 PNG𝕏 分享到 X
#人工智能#代码生成#微软#开发者工具
打开原文

Introducing MAI-Code-1-Flash | Microsoft AI

Skip to main content

Image 1: Microsoft

Source

Our Company

Signal blogOfficial Microsoft BlogMicrosoft On The IssuesAsiaCanadaEurope, Middle East and AfricaLatin AmericaThe Code of UsConexiones

What's new todayAIInnovationDigital TransformationSustainabilitySecurityWork & LifeDiversity & InclusionUnlocked

More

Our CompanySignal blogOfficial Microsoft BlogMicrosoft On The IssuesAsiaCanadaEurope, Middle East and AfricaLatin AmericaThe Code of UsConexiones

What's new todayAIInnovationDigital TransformationSustainabilitySecurityWork & LifeDiversity & InclusionUnlocked

All Microsoft

Software

PCs & Devices

Entertainment

Business

Developer & IT

Other

View Sitemap

Cancel

Search

[](https://microsoft.ai/)

  • [Models](https://microsoft.ai/models/ "Models") ## Latest models

Discover the latest MAI models, designed for real-world intelligence

![Image 2: A watercolor illustration of a thought bubble made up of overlapping pastel circles in pink, yellow, green, and blue, set against a light gray background. ### MAI-Thinking-1](https://microsoft.ai/models/mai-thinking-1/)![Image 3: Two large watercolor curly brackets face each other on a white background. The left bracket is blue and green, while the right bracket is pink and orange. ### MAI-Code-1-Flash](https://microsoft.ai/models/mai-code-1-flash/)![Image 4: Watercolor illustration of two flowers with green stems and leaves. The petals are painted in shades of pink, yellow, orange, and green, set against a plain white background. ### MAI-Image-2.5](https://microsoft.ai/models/mai-image-2-5/)![Image 5: A watercolor illustration of a soundwave with soft, overlapping peaks in green, pink, and yellow tones on a light gray background. ### MAI-Transcribe-1.5](https://microsoft.ai/models/mai-transcribe-1-5/) [All models](https://microsoft.ai/models/ "All models") Image 6: A watercolor illustration of a thought bubble made up of overlapping pastel circles in pink, yellow, green, and blue, set against a light gray background.

[MAI-Thinking-1](https://microsoft.ai/models/mai-thinking-1/)

Built for the complex problems that matter most, with competitive reasoning and top SWE-Bench Pro results at a mid-weight price. Image 7: Two large watercolor curly brackets face each other on a white background. The left bracket is blue and green, while the right bracket is pink and orange.

[MAI-Code-1-Flash](https://microsoft.ai/models/mai-code-1-flash/)

Helps your engineering team write better code faster, with a lightweight, agentic model built into GitHub Copilot and VS Code. Image 8: Watercolor illustration of two flowers with green stems and leaves. The petals are painted in shades of pink, yellow, orange, and green, set against a plain white background.

[MAI-Image-2.5](https://microsoft.ai/models/mai-image-2-5/)

Create stunning, design-ready images from any text or photo prompt, with best-in-class Arena ELO scores at a lower price. Image 9: A watercolor illustration of a soundwave with soft, overlapping peaks in green, pink, and yellow tones on a light gray background.

[MAI-Transcribe-1.5](https://microsoft.ai/models/mai-transcribe-1-5/)

Turn noisy audio into precise, domain-specific transcripts, with leading FLEURS and Artificial Analysis accuracy scores.

  • [Blog](https://microsoft.ai/blog/ "Blog") ## Featured stories

The latest breakthroughs and updates from within our team

Image 10 ## Building a hill-climbing machine:Launching seven new MAI models

7 min read Image 11 ## MAI-Image-2.5 launches at No. 2 for image editing on Arena

4 min read Image 12 ## Introducing MAI-Code-1-Flash

5 min read [All stories](https://microsoft.ai/blog/ "All stories")

  • [Careers](https://microsoft.ai/careers/ "Careers")

[Try Models](https://playground.microsoft.ai/chat "Try Models")

  • [LinkedIn](https://www.linkedin.com/company/microsoft-ai/ "LinkedIn")
  • [Instagram](https://www.instagram.com/microsoftai/ "Instagram")
  • [X](https://www.x.com/MicrosoftAI "X")

Off- [x] Accessibility Mode

Off- [x] Accessibility Mode

Models

Introducing MAI-Code-1-Flash

Introducing MAI-Code-1-Flash

Superintelligence team

June 2, 2026

Models

Superintelligence team

  • [LI](https://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Fmicrosoft.ai%2Fnews%2Fintroducingmai-code-1-flash%2F&title=Introducing+MAI-Code-1-Flash%C2%A0&source=https%3A%2F%2Fmicrosoft.ai%2Fnews%2Fintroducingmai-code-1-flash%2F "Share on LinkedIn")
  • [X](https://x.com/intent/tweet?text=Introducing+MAI-Code-1-Flash%C2%A0&url=https%3A%2F%2Fmicrosoft.ai%2Fnews%2Fintroducingmai-code-1-flash%2F "Share on X")
  • [FB](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Fmicrosoft.ai%2Fnews%2Fintroducingmai-code-1-flash%2F "Share on Facebook")

Today we’re introducing MAI-Code-1-Flash, a new Microsoft coding model built for fast, efficient assistance in everyday developer workflows. It is built end-to-end by Microsoft using clean and appropriately licensed data. The model is rolling out to GitHub Copilot individual users in Visual Studio Code in the model picker and under the default auto picker.

Features and capabilities

  • Agentic coding in real developer environments, trained and designed for GitHub Copilot harness, to work better together.
  • Adaptive thinking, stays concise for simple requests and spends more reasoning budget on complex tasks.
  • Strong instruction-following across single-turn and multi-turn scenarios.

MAI-Code-1-Flash is designed around the simple goal of delivering high-quality coding help with better efficiency. It outperforms Claude Haiku 4.5 with better price to performance across coding benchmarks.

Image 13: A scatter plot compares coding models on pass rate vs. average token usage. MAI-Code-1-Flash (green) outperforms Claude Haiku 4.5 (orange) across benchmarks, with higher pass rates and lower token use in the highlighted “Ideal Zone.”.

Build for developers, not benchmarks

Coding models are most useful when they perform well in the same environment developers use every day. That is why we built MAI-Code-1-Flash with production workflows at the center, rather than optimizing only for benchmarks. The model was trained directly with GitHub Copilot harnesses used in production. This allows it to learn how to interact with surrounding tools and systems in agentic coding tasks, making it uniquely well suited to real-world Copilot workflows compared to other available models.

During training, we evaluated checkpoints across core software engineering tasks, repository question answering, refactoring, and telemetry-grounded tasks adapted from real GitHub Copilot usage. This alignment between training, evaluation, and production helps offline improvements translate into real-world developer quality.

Designed to maximize value per token

MAI-Code-1-Flash was trained with adaptive solution length control, which helps the model adjust the depth of its response to the task. It can stay concise for simpler requests and spend more reasoning budget when a problem requires deeper analysis or broader code changes. In practice, this means developers start seeing useful output sooner. We see MAI-Code-1-Flash solving harder problems with up to 60% fewer tokens. This helps reduce latency, lower cost, improve return on token, and make interactive workflows feel smoother.

Benchmark results in the production harness

To understand both quality and efficiency, we evaluated MAI-Code-1-Flash against Claude Haiku 4.5 on SWE-Bench Verified, SWE-Bench Pro, SWE-Bench Multilingual, and Terminal Bench 2 using the same production harness that developers use for their everyday coding tasks. We measured task success and the average number of solution tokens required to complete each task.

MAI-Code-1-Flash outperforms Claude Haiku 4.5 across all core coding benchmarks tested, with higher pass rates on all 4 evaluations, including a +16-point lead on the diverse, real-world tasks of SWE-Bench Pro (51.2% vs. 35.2%). It’s not just smarter; it’s leaner, solving harder problems with up to 60% fewer tokens on SWE-Bench Verified, proving that higher accuracy and greater efficiency are no longer a trade-off.

Image 14: A comparison table of coding benchmarks for MAI-Code-7-Flash and Claude Haiku 4.5, showing pass rates and average token usage for four benchmarks, with MAI-Code-7-Flash outperforming in all categories.

Math, Science, Instruction Following, and Agentic coding tasks

Image 15: Bar chart comparing four benchmark scores (IF Bench, Advanced IF, Robust IF, τ¹-Bench) for MAI-Code-1-Flash and Claude Haiku 4.5, with MAI-Code-1-Flash consistently scoring higher in all categories.

MAI-Code-1-Flash comes out ahead on every benchmark in the table, with the widest margin on IF Bench precise instruction following (+28.9) and the narrowest on rubric-based Advanced IF (+14.5). The strong instruction-following carries over to agentic tool use.

Furthermore, MAI-Code-1-Flash also outperforms Claude Haiku-4.5 on core reasoning capabilities in math, science, and visual generation coding.

Image 16: A comparison table shows benchmarks for MAI-Code-T-Flash and Claude Haiku 4.5, listing accuracy and average token usage (K) for tasks like math, science, text reasoning, and coding. MAI-Code-T-Flash leads in all benchmarks.

Standard benchmarks reward memorization as much as reasoning, for example a model that has seen the Monty Hall problem will answer it correctly, but invert the prizes and it fails. We built a 186-question, 34-category benchmark around adversarial traps like inverted classics, impossible tasks, and underdetermined scenarios to see whether models were actually reasoning or just pattern-matching. MAI-Code-1-Flash surpasses Claude Haiku 4.5 overall and reached 85.8% adjusted accuracy, with especially strong performance in reasoning, instruction-following, and recognizing impossible problems. We also see room for the model to grow, since core adversarial categories like Einstellung traps remained below 50% accuracy.

Try it out

MAI-Code-1-Flash is now rolling out to VS Code GitHub Copilot individual users. No additional setup is required. As the rollout progresses, you may see GitHub Copilot route tasks to MAI-Code-1-Flash through the Auto picker, or see the model available directly in the model picker.

Here are a few fun sample apps we built with MAI-Code-1-Flash in VS Code:

Video 4

Video 5

We would love to hear from you! Please join the GitHub Community to share your feedback.

Build the Future With Us

Build the Future With Us

We’re a lean, fast-moving lab made up of some of the world’s most talented minds. We have an exciting roadmap of compute at MAI, with our next-generation GB200 cluster now operational. And we have an ambitious mission we truly believe in. We’re also fortunate to partner with incredible product teams giving our models the chance to reach billions of users and create immense positive impact. If you’re a brilliant, highly-ambitious and low ego individual, you’ll fit right in—come and join us as we work on our next generation of models!

Explore all jobs

Related Stories

Related Stories

Image 17

[Building a hill-climbing machine:Launching seven new MAI models](https://microsoft.ai/news/building-a-hillclimbing-machine-launching-seven-new-mai-models/)

announcements

Video 6

Image 18

[MAI-Image-2.5 launches at No. 2 for image editing on Arena](https://microsoft.ai/news/introducing-mai-image-2-5/)

models

Image 19

[Introducing MAI-Voice-2](https://microsoft.ai/news/mai-voice-2expressive-speech-in-10-languages/)

models

[](https://microsoft.ai/)

  • [LinkedIn](https://www.linkedin.com/company/microsoft-ai/ "LinkedIn")
  • [Instagram](https://www.instagram.com/microsoftai/ "Instagram")
  • [X](https://www.x.com/MicrosoftAI "X")

© Microsoft 2026

This site runs Microsoft Clarity for behavioral insights. By using this site, you consent to the collection and use of your data, such as how you interact with this website, by Microsoft for product improvements. Learn more here and at our Privacy Statement.

 English (United States)Your Privacy ChoicesConsumer Health Privacy

What's new

Microsoft Store

Education

Business

Developer & IT

Company

English (United States)

Your Privacy Choices

Consumer Health Privacy

AI 可能会生成不准确的信息，请核实重要内容