Google Rolls Out Gemini 4 Hydrogen and Says the Fast Model Edges Past Argon by One Point

Google has a new favorite word for its cheap, fast model, and it is not Flash. On November 12 the company introduced Gemini 4 Hydrogen, a model it says is built for "rapid execution across diverse agentic workflows," and started rolling it out across AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search and paid tiers of its Antigravity coding tool.

That is a much bigger stage than its heavier siblings got. Earlier this week the company opened Gemini 4 Carbon to paying users. Hydrogen is the version most people will actually touch, because it is going into Search for all users and into the free Gemini app through automatic model selection.

In Google's announcement, Koray Kavukcuoglu, SVP at Google DeepMind and Google's chief AI architect, draws a clean line between the family members. Carbon and Argon remain the "frontier models" for "deep, long-horizon reasoning on your hardest problems." Hydrogen is the worker. It is pitched as a standalone model for everyday jobs, or as the "high-efficiency execution engine" that does the grunt work while a bigger model thinks.

"With Hydrogen, we're also bringing Gemini 4 to everyone." Koray Kavukcuoglu, Google DeepMind

The headline number is a cheeky one. Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking setting, against 50 for Argon, and 54 at a new xhigh setting. One point is not a rout, and it is Google doing the reporting. Still, the fast option claiming to edge past a frontier sibling is clearly the line Google wanted up top.

The price is flat. Google lists Hydrogen at $1.50 per million input tokens and $7.50 per million output tokens across its full 1 million token context window, with up to 1 million output tokens. No long-prompt surcharge, which developers stuffing in whole codebases will notice first.

Google grades its own homework again

Against its predecessor, Gemini 3.8 Flash, Google says Hydrogen finishes tasks 2.5 times faster at its default setting, more than triples Terminal-Bench 4.1 scores (57% against 18.2%) and cuts fabricated claims by more than five times. It says the model leads its class on HLE-Diamond at 58.8%, ranks first overall on the Vals Tax Agent Bench at 52.3%, ties for first on the Vals Legal Research Bench at 55.8% and sits fifth on arena.ai's Agent Arena.

Every one of those figures comes from Google. None is an independent audit, and most comparisons are against Google's own older model or an unnamed "class" of rivals. Treat them as claims to test.

The more interesting idea may be Orchestrator, a new Antigravity feature that switches between Carbon and Hydrogen inside one session. Google says that on Terminal-Bench 4.0, Orchestrator reached "near-Carbon task success rates" while using 24% less quota on average and finishing 19% faster. In parallel setups, Carbon can act as the coordinator while Hydrogen runs worker and verifier agents. That is a frank admission that the expensive model should not be typing every line.

Google is also making safety claims, again measured against 3.8 Flash. It says Hydrogen reduces overreach in agentic tasks by up to 81% and scores a 96.7% honesty rate on what the company calls "our internal agent honesty evaluation." Internal is the key word there.

For regular users, the change will arrive quietly. Google says Hydrogen is "significantly upgrading Search" with more visual answers and a wider range of links, and 9to5Google reports it is now the default model for AI Mode. In the Gemini app, it shows up in the model picker as "Everyday help," while free users only get it when auto selection decides it fits the prompt.

The rollout comes with limits. Google says Hydrogen is "rolling out globally" starting today, which is not the same as live for everyone right now. Antigravity access is limited to paid Google AI Pro and Ultra subscribers, and Hydrogen in Google Sheets starts with those same paid tiers and eligible business plans.