[GitHub Copilot Introduces Multi-Model Collaboration: HydraFusion Cuts Costs by Up to 67%] Beating AI News Flash: GitHub has added a multi-model orchestration system called HydraFusion to Copilot. It first determines how to approach a task, then calls upon different models accordingly. Currently, there are three methods: simple tasks are handled directly by a single model; for complex tasks, a lower-cost model attempts first, and if the result is unsatisfactory, it escalates to a more powerful model; for tasks requiring review, one model completes the task first, another model from a different family identifies errors, and then the first model makes corrections. GitHub compared HydraFusion against Claude Opus 5 on three programming agent benchmarks. It outperformed by 4.9 percentage points on TerminalBench 2.1, underperformed by 1.5 percentage points on DeepSWE, and was nearly on par with a 0.1 percentage point difference on CheckpointBench. However, costs were reduced by 67%, 36%, and 65%, respectively. This approach is reminiscent of Sakana AI's Fugu. Both aim to further refine the question of "which model to choose" into "how to organize multiple models." The difference is that Fugu itself is a trained orchestration model that learns how to call different agents and can even recursively call itself. HydraFusion, on the other hand, currently operates within three fixed execution modes—Single, Cascade, and Critique—making it more akin to embedding multi-model scheduling directly into Copilot. HydraFusion is now available as a research preview for all GitHub Copilot plans and can be enabled through the experimental features in Copilot CLI. GitHub also notes that it is currently best suited for single-turn programming tasks where instructions are clear, while multi-turn, long tasks are still being optimized. [Original Article Link]