GitHub has unveiled Project HydraFusion, a research preview that redefines how developers select AI models for coding tasks. Rather than fixing a single model at configuration time, HydraFusion dynamically constructs an execution plan for each request. For instance, it may draft code with one model, have a second model critique the output, or escalate to a more powerful model if a quality gate rejects the initial attempt. Models are sourced from multiple providers, and developers enable HydraFusion once, much as they would choose any other model.
Is it deployable? Yes, but with limitations. HydraFusion is currently available as a research preview for users on all GitHub Copilot plans, exclusively within GitHub Copilot CLI. There are no open weights or self-hosted options. To enable it, run /update, then /experimental on, then /model and select HydraFusion (Research Preview). Billing is per token consumed by whichever models the workflow invokes, at each model's standard rate.
What the system actually does
HydraFusion builds on GitHub's Auto model selection feature, introduced earlier in 2026 to match tasks to the best-suited model. HydraFusion advances this concept by treating workflow selection as an optimization problem.
It analyzes capability signals for reasoning, code generation, debugging, and tool use, then selects the least complex workflow expected to meet quality standards. This approach minimizes unnecessary model calls, deploying additional resources only when needed to ensure output quality.
By breaking down each task into a multi-step workflow, HydraFusion leverages the strengths of different models across providers—drafting with one, critiquing with another, and escalating only when necessary. This runtime orchestration aims to improve accuracy and efficiency compared to static model selection, potentially reshaping how developers interact with AI-powered coding tools in 2026 and beyond.
via MarkTechPost