Google Antigravity adds Gemini 4 Hydrogen and Orchestrator, which switches between Hydrogen and Carbon mid-session
Google is rolling out Gemini 4 Hydrogen in Antigravity, along with Orchestrator, a model option that hands work back and forth between Hydrogen and Gemini 4 Carbon within one session. Google claims Orchestrator used 24% less quota and finished 19% faster than Carbon on Terminal-Bench 4.1 while reaching near-Carbon task success.
Google is rolling out Gemini 4 Hydrogen inside Google Antigravity alongside Orchestrator, a new model selector option that pairs Hydrogen with Gemini 4 Carbon and switches between them during a single session. The announcement was posted Nov. 12, 2026 on the Antigravity blog.
How Orchestrator works
Google describes Orchestrator as a dynamic, bidirectional approach, in contrast to a static model router that guesses task difficulty from the first prompt and sends the whole task to the cheapest model judged good enough. According to Google, static routers break down as agents run longer and need different levels of complexity at different stages.
Google's example workflow:
- Carbon starts in a read-only phase to explore dependencies, resolve ambiguities and build a plan with explicit re-planning triggers.
- Hydrogen takes over to write edits, update packages and run tests.
- If Hydrogen hits friction, it hands back to Carbon to inspect the state, repair the plan and resume execution.
Google claims that on Terminal-Bench 4.1, Orchestrator reached near-Carbon task success rates while consuming 24% less quota and completing 19% faster on average.
Gemini 4 Hydrogen as a standalone model
Hydrogen can also be selected directly for everyday coding and agentic work. Google reports:
- Artificial Analysis Intelligence Index: 51, ahead of Argon's 50. Google says Hydrogen reaches 54 at its highest setting.
- Terminal-Bench 4.1: 57%, versus 18.2% for 3.8 Flash.
When Hydrogen is selected directly, users can set reasoning depth across four thinking levels: low, medium, high (the default) and a new xhigh tier. Context window, parameter count and API pricing for Hydrogen were not stated in the announcement; pricing not yet disclosed.
Explicit pipelines with the Antigravity SDK
Developers who want to control which model handles each role can wire Hydrogen and Carbon together through the Antigravity SDK. Google's demo is an open-source issue triage system. Hydrogen agents handle triage and coding, including codebase search, clarifying questions and fast UI iteration from PR comments. A Carbon agent is assigned to larger refactors, new features and security checks.
Context: Carbon and /swarm
Google introduced Gemini 4 Carbon earlier this week alongside /swarm, a command for multi-worktree engineering tasks with parallel agent teams. Google says tens of thousands of swarms have been launched on Antigravity in the days since. The company argues that running a frontier reasoning model with multiple parallel agents on every step of routine work slows the loop and burns frontier quota, which is the problem Orchestrator targets.
Availability
- Gemini 4 Hydrogen: rolling out to paid Google AI Pro and Ultra subscribers.
- Orchestrator: rolling out to Google AI Ultra subscribers. Google says additional models and plans will be supported in the future.
- Enterprises: can access both in Gemini for business.
What this means
Orchestrator moves model selection from the user, or from a one-time routing decision at the start of a task, into the course of a single trajectory. Hydrogen can escalate to Carbon when it hits friction, rather than the task being locked to one model. Google's reported Terminal-Bench 4.1 figures (24% less quota, 19% faster, near-Carbon success) are company claims and are tied to that benchmark. They have not been independently verified. For Antigravity users, the practical trade-off Google describes is frontier-level planning at lower quota cost. The SDK route offers the same pairing with manual role assignment. Because Orchestrator is limited to AI Ultra at launch, Pro subscribers have access to Hydrogen directly but not the automatic handoff.
Related Articles
Google launches Gemini 4 Hydrogen: 1M-token context, $1.50 input and $7.50 output per 1M tokens
Google announced Gemini 4 Hydrogen on November 12, 2026, with native 1 million-token input and up to 1 million output tokens. It costs $1.50 per million input tokens and $7.50 per million output tokens, and Google claims it scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level.
Google opens Gemini 4 Carbon to paying developers; API rates double to $4 input, $20 output per 1M tokens on Dec. 1
Google opened Gemini 4 Carbon, its newest reasoning model, to paying developers and enterprises on November 10. Introductory API prices expire November 30, and standard input and output rates double on December 1, according to eWeek.
Perplexity releases Decider v1.3 and Decider Flash v1.3 with text-span extraction and Apache-2.0 weights
Perplexity released Decider v1.3 and Decider Flash v1.3 on November 12, 2026, adding verbatim text-span extraction and downloadable Apache-2.0 weights. The company reports 91.6% mean accuracy for the larger model across 63 query-classification tasks.
Geodesic Intelligence releases NovaAtom-Lite, a biomolecule model it claims tops state-of-the-art systems
Geodesic Intelligence has released NovaAtom-Lite, an AI model that predicts the structures and interactions of biomolecules. Founder Quanquan Gu said the model topped a third-party benchmark in several drug discovery tasks, according to the South China Morning Post.
Comments
Loading...