Gemini 4 Hydrogen reaches Google Search and business AI applications

Google announced the introduction of Gemini 4 Hydrogen into Google Search on November 12, alongside the model's general availability through the Gemini API. The release covers search, the Gemini app and tools for application development. Access and available controls differ between these products: a global Search rollout sits alongside free tiers, API pricing and subscriptions for certain tools used in software development.

In search, Google credits the new model with a more precise understanding of requests containing several conditions, including freely worded questions and typing errors. Responses can include more images, tabular comparisons and tools built for the particular request. The company also says it is improving the selection of links to websites to cover more of the topics involved in a question. These are capabilities described by Google for the new version.

Availability in Romania and differences between features

The official AI Mode availability list includes Romania and Romanian, alongside English. The feature can be accessed from search or the Google app and supports follow-up questions within the same conversation. Individual features have their own availability conditions. For example, interactive visual experiences with simulations and controls for changing variables are described in the current documentation for signed-in users working in English.

The search mechanism runs queries relating to several subtopics in parallel before assembling its response. Google calls this query fan-out. In its documentation for website owners, the company explains that AI Mode and AI Overviews can use different models and techniques. The Hydrogen announcement therefore concerns an update to the search experience, while the two surfaces continue to produce their own responses and link selections.

For websites that can appear among supporting pages, Google's documentation retains the requirement for indexing and eligibility to display a text snippet. The general technical rules require access for Googlebot, an HTTP 200 response and indexable content. Within the same framework, the company explains that participation in AI features uses existing SEO practices, without an additional AI file or a special type of structured data dedicated to those results.

The developer model and its exact limits

Through the API, the public identifier is gemini-4-hydrogen, with general availability status from November 12. Its specification lists text, image, audio, video and PDF inputs, while its output is text. Function calling, structured outputs, code execution and search tools are supported. Image generation and audio generation are outside this model's capabilities, even though Google's interfaces bring several tools together within the same product.

The shared window contains 1,048,576 tokens and is divided between input, reasoning and output. The default limit for synchronous generation is 65,536 tokens. Longer responses use background execution or successive continuations within the remaining available space. The documentation therefore separates total context capacity from the length of a single response; the two values describe different limits applying to the same operation.

The published standard paid-tier price is $1.50 per million input tokens and $7.50 per million output tokens, including reasoning. Input reused through implicit caching has a separate price of $0.075 per million tokens. Searches performed through the Google Search grounding tool are counted separately from the text processed by the model.

The free tier provides Hydrogen access subject to usage limits. The Gemini API terms specify that, for users in the European Economic Area, Switzerland and the United Kingdom, the paid-services data use rules also apply to the free quota and AI Studio. Under that framework, Google does not use prompts and responses to improve products. The terms also require paid services for API applications made available to users in these regions. Romania falls within the scope of these regional conditions.

Reasoning controls and integration changes

Developers can select four reasoning levels: low, medium, high and xhigh, with the last being new. Hydrogen's default API configuration is high. Google describes these levels as a control over processing depth, response time and token consumption. The output limit also includes reasoning, so the budget for that operation covers both intermediate processing and the response delivered to the user.

The migration guide from Gemini 3.8 Flash lists specific changes: Hydrogen does not support explicit context caching, rejects the minimal level and uses a shared window for input and output. Tools and JSON results use native function declarations or response schemas. Preserving conversation state and reasoning signatures remains part of compatibility for sessions involving multiple exchanges.

Multi-model coding and access in the Gemini app

In Google Antigravity, the new Orchestrator option transfers stages of the same task between a model used for planning and Hydrogen for execution. The product description includes handing control back to the planning model when execution encounters an obstacle. The Carbon and Argon versions of Orchestrator are rolling out to Google AI Ultra; Pro subscribers receive the Argon version. Hydrogen is rolling out separately to Pro and Ultra subscribers who select it directly.

In the Gemini app, personal accounts without a subscription use Hydrogen through automatic selection when the system chooses it for the particular prompt. With automatic selection switched off, the documentation specifies Flash-Lite. Accounts with a Google AI plan can manually switch between the models available under their plan. For work or school accounts, the model offering depends on the Workspace licence, a separate criterion from the conditions for personal accounts.

Google positions Hydrogen for everyday coding and agentic workflows, alongside Carbon and Argon for tasks involving extended reasoning. In the Terminal-Bench 4.1 evaluation presented by the company, Hydrogen scored 57%, compared with 18.2% for Gemini 3.8 Flash. Google also reports task completion 2.5 times faster at the default high level in its comparison with Flash. These results are reported by the provider for the specified evaluations.