Unlocking efficiencies in multi-model workloads
In real-world software program growth, whereas groups depend on frontier fashions like Carbon and Argon for deep architectural judgment when designing techniques or untangling delicate bugs, they nonetheless want a quick, disciplined execution mannequin for writing code, working exams, and verifying modifications. Hydrogen is constructed to face by itself as a mannequin for on a regular basis work — or to pair with Carbon and Argon because the high-efficiency execution engine in multi-model workflows.
To assist make this multi-model collaboration automated, we’re introducing Orchestrator in Google Antigravity. Somewhat than utilizing a single mannequin from begin to end, Orchestrator dynamically switches between fashions like Carbon and Hydrogen throughout the identical session. On Terminal-Bench 4.0, Orchestrator reached near-Carbon job success charges whereas, on common, consuming 24% much less quota and finishing 19% sooner. For workflows that require parallel brokers, Carbon can act because the coordinator whereas Hydrogen can energy employee and verifier brokers to construct and check modifications concurrently throughout remoted workspaces — serving to groups ship advanced tasks sooner with out driving up token prices.
