The News, Explained
GitHub announced updated models and workflows for HydraFusion and added HydraFusion Max on October 14, 2026 (October 15 at 05:01 KST). HydraFusion coordinates work across AI models to produce one response. This coordination is called orchestration. Both options remain in research preview across VS Code, the GitHub Copilot app and Copilot CLI, where users work by entering commands. Research preview is a testing stage before behavior is finalized for a general release. Source
The model pool, or set of available models, now includes Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. The models actually used for a task vary with its workflow. Instead of choosing only one model, users select an approach that also coordinates the order of work and review. Source
Changing the review after a draft
Reasoning effort is the level of reasoning applied while solving a problem and reviewing the result. GitHub says HydraFusion previously used only each model’s default effort. It now explicitly uses medium effort for initial drafts and high effort for review, allocating different levels to producing an answer and examining it. Source
Cascade begins with one model and escalates to a stronger model when needed. It can now include a judging stage that inspects the initial work and requests a repair only when necessary. GitHub describes a stronger process for deciding whether additional work is needed. Source
Critique has other models review the completed work. GitHub says the workflow has moved from relying heavily on text summaries toward tools that inspect the actual workspace. It also uses parallel critic models, meaning reviewers work alongside one another. These are GitHub’s descriptions of the process; OYOPICK has not separately measured a reduction in errors. Source
How Max differs and who can use it
| Option | GitHub’s stated design direction |
|---|---|
| HydraFusion | Balance speed, cost and performance, with moderate token savings |
| HydraFusion Max | Apply higher reasoning effort where demanding work needs depth and thoroughness, while considering token efficiency |
Tokens are units used by models to process text. Max uses the same architecture and workflow techniques as ordinary HydraFusion, tuned for work requiring deeper examination. The table compares GitHub’s design goals. The announcement supplies no numerical comparison of response time, accuracy or usage pricing between the options. Source
Both are available on Copilot Pro, Pro+, Business and Enterprise. Experimental or preview features must be enabled, and organization model policies apply. The plan name, enabled-feature requirement and applicable organization policies need to be considered together. Source
OYOPICK’s Take
We find the shift toward inspecting the actual workspace more noteworthy than simply spending longer on a draft. If a review model can examine the original work when its summary differs from the files, that access could help it find and address the discrepancy. That expectation depends on the tools reaching the necessary context and the review leading to appropriate changes.
When choosing between the ordinary option and Max, we think it is more useful to identify the problem that further review should solve than to judge the length of the response. Work involving connections across files or complex requirements gives users more reason to evaluate deeper review. We hope subsequent comparisons report which errors were found and repaired on the same tasks, alongside waiting time and resources used. Such evidence would make it easier to turn Max’s emphasis on depth and the ordinary option’s balance into practical selection criteria.