在受控的离线评估中，HydraFusion 的选择性编码工作流匹配或超过了评估的 Opus 5 基线，同时降低了预估的工作流成本。现已作为 GitHub Copilot 的研究预览提供。
为开发者提供最适合手头任务的模型一直是我们的目标。今年早些时候，我们通过推出自动模型选择（Auto model selection）：https://docs.github.com/copilot/concepts/models/auto-model-selection，使这一过程更加容易，该功能会审查您的任务并匹配最适合该任务的模型。
Learn about artificial intelligence and machine learning across the GitHub ecosystem and the wider industry.
Learn how to build with generative AI.
Change how you work with GitHub Copilot.
Everything developers need to know about LLMs.
Machine learning tips, tricks, and best practices.
Explore the capabilities and benefits of AI code generation and how it can improve your developer experience.
Resources for developers to grow in their skills and careers.
Insights and best practices for building apps.
Tips & tricks to grow as a professional developer.
Improve how you use GitHub at work.
Learn how to move into your first professional role.
Stay current on what’s new (or new again).
Learn how to start building, shipping, and maintaining software with GitHub.
Get an inside look at how we’re building the home for all developers.
Discover how we deliver a performant and highly available experience across the GitHub platform.
Explore best practices for building software at scale with a majority remote team.
Get a glimpse at the technology underlying the world’s leading AI-powered developer platform.
Learn how we build security into everything we do across the developer lifecycle.
Find out what goes into making GitHub the home for all developers.
Our engineering and security teams do some incredible work. Let’s take a look at how we use GitHub to be more productive, build collaboratively, and shift security left.
Explore how to write, build, and deploy enterprise software at scale.
Automating your way to faster and more secure ships.
Guides on continuous integration and delivery.
Tips, tools, and tricks to improve developer collaboration.
DevOps resources for enterprise engineering teams.
How to integrate security into the SDLC.
Learn why Gartner positioned GitHub as a Leader for the second year in a row.
Keep up with what’s new and notable from inside GitHub.
An inside look at news and product updates from GitHub.
The latest on GitHub’s platform, products, and tools.
Insights into the state of open source on GitHub.
The latest policy and regulatory changes in software.
Data-driven insights around the developer ecosystem.
Older news and updates from GitHub.
Learn how to use retrieval-augmented generation (RAG) to capture more insights.
Spotlighting open source maintainers.
How open source is driving positive change.
Explore open source games on GitHub.
Organizations worldwide are incorporating open source methodologies into the way they build and ship their own software.
Stay up to date on everything security.
Demystifying supply chain security.
Updates from the GitHub Security Lab.
Helpful tips on securing web applications.
Learn about core challenges in DevSecOps, and how you can start addressing them with AI and automation.
In controlled offline evaluations, HydraFusion’s selective coding workflows matched or exceeded the evaluated Opus 5 baseline while reducing estimated workflow cost. Now available as a research preview in GitHub Copilot.
Providing developers the best model for the task at hand has always been our goal. Earlier this year, we made that easier by launching Auto model selection,：https://docs.github.com/copilot/concepts/models/auto-model-selection which reviews your task and matches it to the best-suited model for that task.
Today, we’re introducing Project HydraFusion, a research preview that delivers frontier intelligence through runtime orchestration. It creates a full execution plan, choosing from models across multiple providers to draft, critique and revise, or cascade to more powerful models to complete your task.
HydraFusion fills a key role in our overall strategy to deliver automated semantic routing between local, cloud, and compound models. For developers, that complexity stays behind the scenes: you select HydraFusion like any other model, and it chooses a workflow that balances performance, cost, and latency for each task.
HydraFusion is available to users on all GitHub Copilot plans through /experimental in GitHub Copilot CLI：https://docs.github.com/copilot/how-tos/copilot-cli/use-copilot-cli/overview. Usage is based on the tokens consumed by the models HydraFusion uses, priced at each model’s standard rate：https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing.
Please post feedback in the GitHub Community：https://github.com/orgs/community/discussions/206492.
HydraFusion treats workflow selection as an optimization problem. It uses capability signals for reasoning, code generation, debugging, and tool use to select the most efficient execution pattern to meet the quality bar.
For each request, HydraFusion currently chooses one of three execution patterns:
Each pattern addresses a different quality-to-cost trade-off. Single preserves speed and efficiency when one model can solve the task directly. Cascade gives an efficient model the first attempt while retaining a path to stronger inference when the candidate does not clear the acceptance gate. Critique adds an independent perspective for tasks where review is more useful than another unaided attempt.
In offline evaluations across three agentic coding benchmarks, HydraFusion consistently demonstrated frontier-level quality with substantial estimated cost savings. On TerminalBench 2.1, it improved verified task quality by 4.9 percentage points at 67% lower estimated cost compared with Claude Opus 5.
Let’s dive into the approach, the results, and the benchmarks.
Developers already coordinate models manually: choosing one for a task, asking another to review the work, or escalating a difficult problem to a more capable model. HydraFusion brings that familiar process into the runtime. You choose HydraFusion once and stay focused on your task while it manages the models and workflow behind the scenes.
The key is selectivity. Some coding tasks can be solved directly, while others benefit from review, revision, or escalation. HydraFusion evaluates each request and chooses the least complex workflow expected to meet its needs, using additional model calls only when they are likely to improve the result. This adaptive approach balances quality, cost, and latency across models.
As the model frontier advances, so does HydraFusion. When new models become available in GitHub Copilot, we can evaluate and incorporate them into its model pool, bringing their strengths to the tasks best suited to them.
Turning adaptive multi-model orchestration into one dependable coding experience requires careful control of execution, review, cost, and repository state. HydraFusion is built around five operating principles:
Together, these principles make multi-model orchestration practical for repository-level work. Internally, the runtime records the role, outcome, cost, latency, and diagnostics of each leg so the workflow can be understood after execution. Externally, the developer receives one coherent response and one permission-aware change set.
扫码直达文章详情