Gloo Code claims 70% Terminal-Bench at lower cost

Gloo launched Gloo Code, reporting approximately 70% on Terminal-Bench 2.1 at 58%–67% lower cost than published results for Gemini 3 Pro, Sonnet 5, and OpenAI 5.6 Luna. In its official launch release, Gloo says the benchmark is preliminary, internally tested, and compared against third-party published results; 500+ builders get access through its 30-day hackathon. That caveat makes GitHub HydraFusion’s 67% saving that still needs a billing test useful context: treat routing and harness savings as a procurement hypothesis until your own tasks and invoices reproduce them.