GitHub released a research preview called Project HydraFusion that automatically selects and coordinates multiple AI models for a coding task instead of relying on one, according to GitHub's own announcement.
The system picks from three patterns depending on the task: single, where one model solves it directly; cascade, where an efficient model drafts a solution that escalates to a stronger model only if it fails a quality check; and critique, where one model drafts, a second independently reviews, and the first revises based on that feedback. "You select HydraFusion like any other model, and it chooses a workflow that balances performance, cost, and latency for each task," GitHub said.
Across three benchmarks, GitHub reported mixed but mostly favorable results against using Claude Opus 5 alone: 4.9 percentage points better quality at 67% lower cost on TerminalBench 2.1, 1.5 points lower quality but 36% cheaper on DeepSWE, and nearly identical quality at 65% lower cost on CheckpointBench. The model pool includes Claude Opus 5 and GPT-5.6 Sol, GitHub said.
The cost-versus-quality call on which model handles a given request is quietly becoming something the tool decides for you, not something a developer picks up front.