HydraFusion 更新并推出全新 Max 选项
英文标题：HydraFusion updates and new Max option
GitHub 更新了 HydraFusion 的模型池和编排工作流，并推出面向高难度任务的 HydraFusion Max；两者目前仍处于研究预览阶段。新模型池包含 Claude Opus 5.5、GPT-6 Astra、GPT-6.1 Sol 和 GPT-6 Luna，系统会根据编排工作流为不同任务选择模型。更新还针对不同步骤调整推理强度，并优化了 Cascade 与 Critique 工作流。开发者可在 VS Code、GitHub Copilot 应用和 GitHub Copilot CLI 中使用这两项能力，适用于启用实验性或预览功能的 Copilot Pro、Pro+、Business 和 Enterprise 套餐。
12.2 KB
26.4 KB
中文
HydraFusion 更新了模型池和编排工作流，并推出 HydraFusion Max。HydraFusion 和 HydraFusion Max 均可在 VS Code、GitHub Copilot 应用和 GitHub Copilot CLI 中使用，目前仍处于研究预览阶段。
HydraFusion 会编排多个模型以生成单一响应，整合各模型的优势，使你无需为每项任务手动选择单个模型。
- **最新前沿模型：**HydraFusion 和 HydraFusion Max 使用的模型池现已纳入 Claude Opus 5.5、GPT-6 Astra、GPT-6.1 Sol 和 GPT-6 Luna。每项任务所使用的模型会根据编排工作流而变化。
- **针对每个步骤定制推理强度：**此前，HydraFusion 仅使用各模型默认的推理强度。通过此次更新，它会明确为每个阶段定制推理强度：初稿阶段使用中等强度，审查阶段使用高强度。这样既能保持起草效率，也能将更深入的推理投入到检查和改进工作中。
- 优化 Cascade 和 Critique 工作流：
- **Cascade：**此前会使用初始模型解决任务，并在需要时升级至更强模型。现在，该流程可加入评判步骤，检查初始工作，并仅在必要时请求修复。
- **Critique：**此前高度依赖文本摘要。现在，该流程会更多地借助工具直接检查实际工作区。Critique 现在还会使用并行的批评模型来检查工作成果。
HydraFusion Max 基于与 HydraFusion 相同的架构，并采用相同的工作流技术，但针对要求更高的任务进行了前沿能力调优。它会在深度和全面性最重要的环节使用更高的推理强度，同时仍通过编排机制兼顾 Token 效率。
| 选项 | 说明 | 典型使用场景 |
|---|---|---|
| HydraFusion | 采用均衡方案，并实现适度的 Token 节省。 | 需要在速度、成本和性能之间取得平衡的 Agent 任务。 |
| HydraFusion Max | 面向要求更高任务的前沿能力，并通过编排兼顾 Token 效率。 | 需要尽可能深入推理和最全面分析的任务。 |
两个选项均支持上述三个客户端，因此无论你是在 VS Code、GitHub Copilot 应用还是终端中工作，都可以使用最新的 HydraFusion 体验。
HydraFusion 和 HydraFusion Max 面向启用了实验性或预览功能的 Copilot Pro、Pro+、Business 和 Enterprise 套餐提供，但需遵守组织的模型策略。
请参阅 HydraFusion 文档了解更多信息。
欢迎在 HydraFusion 讨论区分享反馈。
English / 原文
HydraFusion has an updated model pool, improved orchestration workflows, and introduces HydraFusion Max. Both HydraFusion and HydraFusion Max are available across VS Code, the GitHub Copilot app, and GitHub Copilot CLI and remain in research preview.
HydraFusion orchestrates multiple models to produce a single response, bringing together their strengths so you don’t have to choose an individual model for each task.
- Latest frontier models: HydraFusion and HydraFusion Max draw from a pool that now includes Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. The models used for each task vary based on the orchestration workflow.
- Reasoning effort tailored to each step: Previously, HydraFusion only used each model’s default reasoning effort. With this update, it explicitly tailors effort to each stage, using medium effort for initial drafts and high effort for review. This keeps drafting efficient while dedicating deeper reasoning to checking and improving the work.
- Optimized Cascade and Critique workflows:
- Cascade: Previously used an initial model to solve the task, with escalation to a stronger model when needed. It can now include a judging step that inspects the initial work and requests a repair only when needed.
- Critique: Previously relied heavily on text summaries. It now relies more on tools to inspect the actual workspace directly. Critique now also uses parallel critic models to inspect work.
HydraFusion Max is built on the same architecture and uses the same workflow techniques as HydraFusion, but is tuned for frontier capability on more demanding tasks. It applies higher reasoning efforts where depth and thoroughness matter most, while still using orchestration to consider token efficiency.
| Option | Description | Typical use case |
|---|---|---|
| HydraFusion | Balanced approach with moderate token savings. | Agentic tasks that require a balance of speed, cost, and performance. |
| HydraFusion Max | Frontier capability for more demanding tasks, with token efficiency considered through orchestration. | Tasks requiring the deepest possible reasoning and most thorough analysis. |
Both options are supported across all three clients, so you can access the latest HydraFusion experience whether you work in VS Code, the GitHub Copilot app, or the terminal.
HydraFusion and HydraFusion Max are available on Copilot Pro, Pro+, Business, and Enterprise plans with experimental or preview features enabled, subject to your organization’s model policies.
Learn more in the HydraFusion documentation.
Share your feedback in the HydraFusion discussion.