Google on Thursday announced the global rollout of Gemini 4 Hydrogen, its newest AI model, moving it beyond internal testing to all users.

The model is live in Google Search and selectable in the Gemini app's model dropdown, replacing Gemini 3.8 Flash, which launched in September, as the engine behind Search's AI features.

DeepMind SVP Koray Kavukcuoglu described Hydrogen as built for 'rapid execution across diverse agentic workflows' — tasks like writing code, running tests, and handling everyday work — rather than the deep, long-horizon reasoning that frontier models Carbon and Argon handle.

Availability also extends to Google AI Studio, the Gemini API, Android Studio's paid tier, Google Antigravity, and Gemini in Google Sheets, with Pro and Ultra subscribers getting access in enterprise surfaces.

The surprising result is that a speed-oriented model outscored Google's own frontier release.

At its default 'high' thinking level, Hydrogen hit 51 on the Artificial Analysis Intelligence Index against Argon's 50, and reached 54 at the new xhigh setting, according to Google's announcement as reported by TradingView.

That means Google's latest fast model edges the frontier model it shipped just weeks ago on a general intelligence benchmark — an unusual inversion.

Google also claims Hydrogen completes tasks 2.5 times faster than predecessor Gemini 3.8 Flash, and that it is two times less likely than 3.8 Flash to circumvent permission boundaries. All benchmark figures come from Google's own claims via the pack's reports.

Hydrogen introduces a four-tier thinking system — low, medium, high (the default), and a new xhigh tier — that varies in quality, latency, and cost, letting simpler requests run cheaper while complex ones get more reasoning.

Google is pricing the model at $1.50 per million input tokens and $7.50 per million output tokens, per TradingView.

Google claims the tiers are designed to finish tasks faster while consuming fewer output tokens, reflecting Hydrogen's core pitch: not every request needs the most computationally demanding model.

Google DeepMind says Hydrogen can 'stand on its own as a model for everyday work — or pair with Carbon and Argon as the high-efficiency execution engine in multi-model workflows.'

For Search users, Tech Times reports the upgrade is the biggest model improvement to Search yet. Hydrogen is built to parse multi-part questions, understand typos, and follow users' trains of thought, so people can state several conditions in one query instead of splitting it into separate searches.

It also improves Google's search fan-out technique, breaking one question into multiple subqueries to surface websites that a narrow interpretation might miss. AI Mode responses become richer too, with more images, comparison tables, structured information, and custom tools.

The company cautions that Hydrogen remains prone to hallucinations like all generative AI models, and that jailbreak-resistance work is ongoing.

The announcement landed well with investors: TradingView reports Alphabet shares rose 2% in midday trading after the launch.

On agentic benchmarks, Google says Hydrogen ranks fifth on Agent Arena across multi-step work such as deep research, data analysis, and navigating large codebases — solid for an execution model, though it is not positioned to replace Carbon or Argon for architectural judgment.

With the model available at every tier from free Search users to Ultra subscribers, adoption metrics and independent benchmark verification will be the next things to watch.