Gemini 4 Hydrogen is Google's fast Gemini 4 model, and Antigravity pairs it with Argon or Carbon

Google released Gemini 4 Hydrogen on 12 November, a faster and cheaper member of the Gemini 4 family built for everyday coding and agent work. Koray Kavukcuoglu’s Google Blog post says it is rolling out in the Gemini API (AI Studio and Android Studio), Gemini for business, the Gemini app, Google Search and Google Antigravity. The price is flat across its 1 million-token context window: $1.50 per million input tokens and $7.50 per million output tokens.

Where Hydrogen sits

Google positions Hydrogen below Gemini 4 Argon and Carbon, as the “fast, disciplined execution model” that writes code, runs tests and checks changes while a bigger model plans. It is also pitched as the upgrade path from Gemini 3.8 Flash.

Google’s own numbers, all unverified by us:

• Intelligence Index. 51 at the default High thinking level, against Argon’s 50, and 54 at the highest setting. The Artificial Analysis release page lists the Xhigh configuration at 54.
• Terminal-Bench 4.1. 57%, against 18.2% for Gemini 3.8 Flash.
• Speed. About 2.5 times faster than 3.8 Flash at the default setting, and fewer fabricated claims, by Google’s count “more than 5x”.

In the Gemini app, Google says free users get Hydrogen through automatic model selection when it fits the prompt.

Orchestrator: a planner plus Hydrogen

The bigger product change is in Antigravity. A new model-selector option, Orchestrator, pairs a planner with Hydrogen inside one conversation. The models docs describe a read-only planning phase, then Hydrogen edits files, builds and tests. If Hydrogen hits friction, such as a verification step that fails twice, control goes back to the planner, which repairs the plan and hands execution back. The docs say this is not a router that classifies the first prompt; the hand-offs happen within a single response. You cannot pick a thinking level for Orchestrator.

There are two versions. Orchestrator (Carbon) plans with Gemini 4 Carbon and is Google AI Ultra only. Orchestrator (Argon) plans with Argon and is listed for Ultra and for paid Google AI Pro plans (or trials on eligible premium devices). Neither is listed for Free, Plus or Enterprise. Hydrogen itself is listed on every plan, including Free and Plus. The changelog dates the launch 12 November and notes that new versions roll out gradually. We have not signed in to confirm access on any plan.

Google says Orchestrator reached near-Carbon task success on Terminal-Bench while using 24% less quota and finishing 19% faster on average. The Antigravity post attributes that to Terminal-Bench 4.1; the Google Blog says Terminal-Bench 4.0. Neither publishes Carbon’s standalone score for comparison.

In the CLI, the extra reasoning level is xhigh, for example --model gemini-4-hydrogen-xhigh. This builds on Monday’s /swarm preview and Carbon launch; Orchestrator keeps one agent, while /swarm runs many.

Why this matters

Hydrogen gives Google a cheaper Gemini 4 option for the high-volume part of agent work, at a flat price across a 1 million-token window. Orchestrator turns the “big model plans, small model types” pattern into a menu choice, and the Argon version brings it to Pro subscribers rather than only Ultra. The quota and speed savings are Google’s claims, and the two official posts disagree on which Terminal-Bench version produced them.