GitHub 使 Copilot 聚变在2026宇宙中一般可用

AI 模型和平台

GitHub 使 Copilot 聚变在2026宇宙中一般可用

在 Google 上的首选来源中添加 Unite.AI

GitHub制造了Copilot Fusion，其多型号的路由和管弦系统，一般于2026年10月28日启用，在GitHub宇宙2026与新的GitHub Copilot应用，Copilot项目并列揭幕，更深一层的Copilot代码审查级。

这些公告来自CoreAI执行副总裁Jay Parikh在Microsoft的2026 University主旨文章，其团队包括GitHub，以及一系列同日的更改日志条目，详细介绍了可用性，计费和预览术语。

Copilot融合路线和管弦乐模型

在模型拾取器中选择Copilot Fusion，根据Copilot Fusion changlog选择一个路由和管弦系统，而不是特定的模型。 对于每个提示，Fusion决定使用哪种模型以及它们如何合作，将工程Hydrafusion的多模式管弦与Copilot的汽车模型选择特征背后的微调模型选择相结合。

用户从三个工作级别中选择一个。 Fusion Standard优先处理日常编码的多功能模型，Fusion High针对挑战性的任务和更深层次的推理，Fusion Max为最艰巨的任务和复杂的推理保留了最有能力的模型。 对于每项任务，Fusion运行三个执行工作流程中的一个：单体，一个模型直接解决任务；连体，一个高效的模型起草一个响应，一个质量检查可以接受它或者升级为更强的模型；Critique，一个模型起草，第二个模型审查，原模型修改。

GitHub 称 Copilot Fusion Max在人工分析编码Agent指数上得分60分，与Opus 5.5 Max在Claude Code的两个点内，每任务的平均API成本约为一半，使用约一半的符号，每任务约减少40%的时间。

管理者通过企业和组织模式政策来控制哪些模式可以借鉴Fusion，他们可以禁用模式用于Fusion，而不会令每个人失去功能。 企业管理设置还可以设置Fusion，在任何努力级别，作为Copilot用户的默认模式。

Hari Pokhrel,AT&T公司的主要软件工程师，称在变换日志中称，Fusion Standard通过自动为每项任务选择正确的模型，减少不必要的开支并保持高价值工作的能力，提高了整个AT&T的象征性效率，其决策也随着时间的推移得到了改进。

Fusion按每个模型的标准费率向每个模型计费，付费的Copilot计划中的用户在使用时会获得有限时间的30%的模型成本折扣。 该系统在Visual Studio Code,GitHub Copilot app,GitHub Copilot CLI,JetBrains IDEs等所有Copilot计划上均可使用，并有额外的客户端支持。 在研究预览期间启用自动，Hydrafusion，或Hydrafusion Max的开发者会自动过渡到相应的Fusion努力级别。

GitHub Copilot App和Copilot项目

GitHub Copilot app是为代理软件的发送而设计的，Parikh写道，将agents和工具连接到各个项目，并支持超出代码的工作，包括写谱，介绍设计讨论，以及在一个表面分析遥测。

Copilot项目，自2026年10月28日起在app中本土提供，让用户运行agent在全管理，平行的会话中跨多个寄存器工作。 Copilot app changlog 标签 Copilot Projects a public preview：用户描述一个目标，附加诸如寄存器，设计模拟，或 Slack 线程等上下文，选择模型和工具，并接收拉动请求审查等结果。 项目自动化处理可重复的任务，例如每当一个功能获得时都会创建宣传视频。

该app还可以使用Slack和Figma等连接工具的上下文，可以在聊天时从自然语言生成和编辑图像。 一个计算机使用特性，在macOS和Windows的公开预览中，让Copilot读取应用软件内容，点击控制，输入或编辑文本，并浏览过桌面应用程序，包括没有API，命令行界面，或MCP整合的应用程序。 Copilot在控制一个应用程序之前请求批准，组织可以通过管理设置来禁用该特性。

会话连续性让用户在GitHub应用，VS代码，CLI，和网络上开始并跟踪同样的云 agent会话。

代码审查 Max、 自动化和企业访问

GitHub同样是推出 Copilot代码审查马克斯，Parikh 称在Review Bench上贴出了最高的F1得分，用于高热度问题。 现有的 " 文学与平衡 " 级仍供日常审查，" 马克斯 " 级意在需要更深入审视的变革。

Copilot自动化器在一个在云中运行，手动触发，由事件，或按期运行的存储器内创建AI动力工作流程，用于发行分解和代码审查等任务。 GitHub 称自动化机现在可以学习其成果，并提出改进建议供用户审查和批准。

从2026年10月28日开始，GitHub Entertainment 客户可以使用所有Copilot特性，包括Copilot的app，代码审查，以及自动化，而无需单独购买订阅，代之以支付他们使用的AI信用。 通过GitHub Sales即刻提供这一变化，自服务器即将到来；现有的Copilot Business和Copilot Entertainment 客户可以保留其目前的计划或待准备后搬走。

从海合金和基础设施增加的道路

Fusion的一般可用性遵循了工程Hydrafusion研究预览GitHub 推出 2026年9月4日，在当年早些时候的发布汽车模型选择之后。 在预览版报告的离线评价中，固定的海合金政策提高了经核实的任务质量4.9个百分点，比Bench 2.1号终端的Claude Opus 5的估计费用低67%，在DeepSWE的1.5个百分点内低36%，在Bench检查站的0.1个百分点内低65%的费用，GitHub的内部基准测试由真正的Copilot会议建造。 HydraFusion的三个执行模式——Single,Cascade，和Critique——传入Fusion的工作流程集。

GitHub 称在2026年至今增加了400万个CPU核心，160个PB高速存储，50个Tb/s网络容量。 3项原则指导其新的Git架构：简化原始体，解密存储与计算，尽量减少协调。 该公司正在迁移到Azure的多地区部署，预计将在2027年初完成这一迁移。

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。

AI Models & Platforms

GitHub Makes Copilot Fusion Generally Available at Universe 2026

Add Unite.AI to your preferred sources on Google

GitHub made Copilot Fusion, its multi-model routing and orchestration system, generally available on October 28, 2026, unveiling it at GitHub Universe 2026 alongside the new GitHub Copilot app, Copilot projects, and a deeper Copilot code review tier.

The announcements arrived in a Universe 2026 keynote post from Jay Parikh, Executive Vice President of CoreAI at Microsoft, whose teams include GitHub, and in a series of same-day changelog entries detailing availability, billing, and preview terms.

How Copilot Fusion Routes and Orchestrates Models

Selecting Copilot Fusion in the model picker chooses a routing and orchestration system rather than a specific model, according to the Copilot Fusion changelog. For each prompt, Fusion determines which models to use and how they work together, combining the multi-model orchestration of Project HydraFusion with the fine-tuned model selection behind Copilot’s auto model selection feature.

Users pick one of three effort levels. Fusion Standard prioritizes versatile models for everyday coding, Fusion High targets challenging tasks and deeper reasoning, and Fusion Max reserves the most capable models for the most demanding tasks and complex reasoning. For each task, Fusion runs one of three execution workflows: Single, where one model solves the task directly; Cascade, where an efficient model drafts a response and a quality check accepts it or escalates to a stronger model; and Critique, where one model drafts, a second model reviews, and the original model revises.

GitHub said Copilot Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the average API cost per task, using about half as many tokens and about 40% less time per task.

Administrators control which models Fusion can draw on through enterprise and organization model policies, and they can disable models for Fusion without disabling them for everyone. Enterprise managed settings can also set Fusion, at any effort level, as the default model for Copilot users.

Hari Pokhrel, a principal software engineer at AT&T, said in the changelog that Fusion Standard improved token efficiency across AT&T by automatically selecting the right model for each task, reducing unnecessary spend and preserving capacity for higher-value work, and that its decision-making has improved over time.

Fusion is billed per model at each model’s standard rate, and users on paid Copilot plans receive a limited-time 30% discount on model costs when using it. The system is available on all Copilot plans in Visual Studio Code, the GitHub Copilot app, GitHub Copilot CLI, and JetBrains IDEs, with additional client support to come. Developers who enabled auto, HydraFusion, or HydraFusion Max during the research preview will transition automatically to the corresponding Fusion effort levels.

The GitHub Copilot App and Copilot Projects

The GitHub Copilot app is designed for agentic software delivery, Parikh wrote, connecting agents and tools across projects and supporting work beyond code, including writing specs, presenting design discussions, and analyzing telemetry in one surface.

Copilot projects, available natively in the app starting October 28, 2026, let users run agent work across multiple repositories in fully managed, parallel sessions. The Copilot app changelog labels Copilot Projects a public preview: users describe a goal, attach context such as repositories, design mocks, or Slack threads, choose models and tools, and receive results such as pull requests for review. Project automations handle repeatable tasks, such as creating a promo video whenever a feature lands.

The app can also use context from connected tools such as Slack and Figma, and it can generate and edit images from natural language in chat. A computer-use feature, in public preview on macOS and Windows, lets Copilot read app content, click controls, enter or edit text, and navigate across desktop apps, including those without an API, command-line interface, or MCP integration. Copilot asks for approval before controlling an app, and organizations can disable the feature through managed settings.

Session continuity lets users start and follow the same cloud agent session across the GitHub app, VS Code, the CLI, and the web.

Code Review Max, Automations, and Enterprise Access

GitHub also introduced Copilot code review Max, which Parikh said posted the highest F1 score on ReviewBench for high-severity issues. The existing Lite and Balanced tiers remain for everyday reviews, with Max intended for changes that need a deeper look.

Copilot automations create AI-powered workflows inside a repository that run in the cloud, triggered manually, by events, or on a schedule, for tasks such as issue triage and code review. GitHub said automations can now learn from their outcomes and propose improvements for users to review and approve.

Starting October 28, 2026, GitHub Enterprise customers can use all Copilot features, including the Copilot app, code review, and automations, without buying a separate subscription, paying instead for the AI credits they use. The change is available immediately through GitHub Sales with self-serve coming soon; existing Copilot Business and Copilot Enterprise customers can keep their current plans or move over when ready.

The Road From HydraFusion and Infrastructure Additions

Fusion’s general availability follows the Project HydraFusion research preview GitHub introduced on September 4, 2026, after launching auto model selection earlier in the year. In offline evaluations reported with the preview, fixed HydraFusion policies improved verified task quality by 4.9 percentage points at 67% lower estimated cost than Claude Opus 5 on TerminalBench 2.1, came within 1.5 percentage points at 36% lower cost on DeepSWE, and came within 0.1 percentage points at 65% lower cost on CheckpointBench, GitHub’s internal benchmark built from real Copilot sessions. HydraFusion’s three execution patterns — Single, Cascade, and Critique — carry into Fusion’s workflow set.

GitHub said it has added 4 million CPU cores, 160 PB of high-speed storage, and 50 Tb/s of network capacity so far in 2026. Three principles guide its new Git architecture: simplify the primitives, decouple storage from compute, and minimize coordination. The company is migrating to a multi-region deployment in Azure and expects to complete that migration in early 2027.

Auto-fetched and translated by AI Atlas. All rights belong to the original author.