https://beanhsiang.github.io/tags/AI-Agents/ Recent content in AI Agents on Innovation with tech Hugo zh-Hans Sat, 03 Oct 2026 12:00:00 +0800 https://beanhsiang.github.io/post/2026-10-03-hydrafusion-deep-dive-multi-model-orchestration-in-github-copilot/ Sat, 03 Oct 2026 12:00:00 +0800 https://beanhsiang.github.io/post/2026-10-03-hydrafusion-deep-dive-multi-model-orchestration-in-github-copilot/ <p> <a data-fancybox="gallery" href="https://beanhsiang.github.io/uploads/20261003161935-b6ce6421d9b945eaad5a7a0edd63dc86.jpg"> <img class="mx-auto" alt="Research preview: HydraFusion. Advanced runtime model orchestration in GitHub Copilot." src="https://beanhsiang.github.io/uploads/20261003161935-b6ce6421d9b945eaad5a7a0edd63dc86.jpg" /> </a> </p> <p><em>题图来自 GitHub Blog 的 <a href="https://github.blog/ai-and-ml/github-copilot/project-hydrafusion-frontier-quality-via-multi-model-orchestration/?WT.mc_id=AI-MVP-5003172">Project HydraFusion</a>。</em></p> <p>9 月初 GitHub 放出了 HydraFusion，一个在 Copilot 模型选择器里排着的&quot;模型&quot;，但它其实不是模型。我把官方博客、Changelog、社区讨论帖、HyDRA 论文和 Rubber Duck 的两篇文章放在一起读了一遍，想弄清楚三件事：它到底在每一轮里做了什么，那组&quot;成本降 67%、质量还更高&quot;的数字能信到什么程度，以及现在值不值得切过去用。</p>