Token & cost optimizationImpact: Medium
Oleksandr Kuzmenko, EditorPublished 6 min readAI-assisted · editor-reviewed
TL;DR
- Orchestrator alternates between Gemini 4 Carbon for reasoning and Hydrogen for code execution.
- Terminal-Bench 4.1 tests show a 24% quota reduction and 19% speed improvement compared to Carbon alone.
- Antigravity SDK enables custom agent pipelines pairing Hydrogen sub-agents with Carbon supervisors.
Key facts
- Quota reduction vs Carbon
- 24% less quota (self-reported)
- Execution latency vs Carbon
- 19% faster (self-reported)
- Standalone Hydrogen Terminal-Bench 4.1
- 57% (self-reported)
- Standalone Hydrogen Artificial Analysis Index
- 51 (default), up to 54 at highest setting (self-reported)
Bidirectional Multi-Model Trajectories. Static model routing typically assesses difficulty only on the first user prompt and routes the entire job to one tier. Orchestrator replaces this with dynamic, bidirectional handoffs between Gemini 4 Carbon and Gemini 4 Hydrogen within a single agent trajectory. Carbon executes an initial read-only exploration phase to map repository dependencies and generate execution plans with explicit replanning triggers. Hydrogen then performs concrete file edits, updates manifests, and runs test suites. If Hydrogen encounters unexpected execution failures, Orchestrator returns state control to Carbon to diagnose the diff, update the plan, and hand execution back to Hydrogen.
Antigravity SDK and Benchmark Metrics. On Terminal-Bench 4.1 benchmarks, Google reports that Orchestrator achieves near-Carbon success rates while completing tasks 19% faster and consuming 24% less quota. Developers can construct programmatic multi-agent pipelines with the Antigravity SDK, deploying multiple Hydrogen sub-agents for issue triage and UI iteration while reserving Carbon for complex refactoring and security audits. Gemini 4 Hydrogen is available to Google AI Pro and Ultra tiers with configurable reasoning settings (low, medium, high, xhigh), whereas Orchestrator is currently restricted to Google AI Ultra subscribers.
When to use
- Use Orchestrator for complex, multi-step engineering tasks requiring exploratory planning followed by repetitive code modifications.
- Use standalone Hydrogen with xhigh reasoning when you need high-speed execution without burning Carbon token allowances.
When NOT to use
- Do not use Orchestrator if your workspace requires strictly deterministic single-model outputs during compliance runs.
- Avoid using Orchestrator on Google AI Pro plans, as it currently requires a Google AI Ultra subscription.
What to do today
- Select Orchestrator in the Antigravity model dropdown when running full-repository refactoring sessions.
- Adjust standalone Gemini 4 Hydrogen reasoning depth to xhigh for edge-case debugging tasks.
- Configure custom multi-agent routing using the Antigravity SDK to offload triage tasks from Carbon.
Step names are taken from What to do today.
Step names are taken from What to do today. Select Orchestrator in the Antigravity model dropdown when running full-repository refactoring sessions.. Adjust standalone Gemini 4 Hydrogen reasoning depth to xhigh for edge-case debugging tasks.. Configure custom multi-agent routing using the Antigravity SDK to offload triage tasks from Carbon.
- Select Orchestrator in the Antigravity model dropdown when running full-repository refactoring sessions.
- Adjust standalone Gemini 4 Hydrogen reasoning depth to xhigh for edge-case debugging tasks.
- Configure custom multi-agent routing using the Antigravity SDK to offload triage tasks from Carbon.
What the community says
Before I go and burn through my Pro tokens in four minutes again (looking at you, Argon): is this something usable with a mere Pro account? Or will this require Ultimate for agy?
Thing is Carbon should be in the middle by this logic. But no - it is their heavyweight model.
#Google Antigravity#Gemini 4 Hydrogen#Gemini 4 Carbon#Antigravity SDK
