GitHub has announced Project HydraFusion, a research preview that combines multiple models at runtime. This feature creates the optimal execution plan for a given task, selecting from models provided by multiple providers to perform drafting, critical review, correction, or delegation to more powerful models.
Product LaunchesGitHubProject HydraFusion
GitHub Copilot Unveils Project HydraFusion, Dynamically Combining Multiple Models
This article is a translation. Read the Japanese original
Developers can use it simply by selecting HydraFusion just like a standard model, without needing to be conscious of the complex underlying processes. The system determines the optimal workflow by considering the balance between performance, cost, and latency.
There are currently three execution patterns: "Single," which attempts to solve the task with a single model; "Cascade," which tries with an efficient model and connects to a powerful model if necessary; and "Critique," which seeks a review from a different model. These are used to optimize the trade-off between quality and cost.
In offline evaluations, GitHub reported that it demonstrated state-of-the-art quality in agentic coding benchmarks. For example, they stated that in TerminalBench 2.1, it improved task quality by 4.9% while reducing estimated costs by 67% compared to Claude Opus 5.
It is available for trial by all GitHub Copilot plan users through the /experimental command in the GitHub Copilot CLI. Usage is billed based on token consumption at the standard rates of each model used.
Source: Project HydraFusion: Frontier quality via multi-model orchestration (Hacker News Frontpage, 2026-09-05)