Gemini 4 Hydrogen

Gemini 4 Hydrogen is Google’s model for everyday coding, file analysis and tasks that require an AI system to use tools. Google introduced it on 12 November 2026, positioning it alongside Carbon and Argon for different levels of work.

Input and output share one budget

The developer documentation sets a combined input and output limit of 1,048,576 tokens. Hydrogen accepts text, images, video, audio and PDFs, and returns text. A token is a piece of encoded content, rather than a word.

The default synchronous output limit is 65,536 tokens. Long decode can produce up to 1,048,576 output tokens, but that output still shares the overall budget with the input. A request using a large amount of input therefore has less room available for output. Developers should budget for both sides of the exchange.

API prices are charged per million tokens

Google’s pricing page lists standard rates of $1.50 for input, $7.50 for output and $0.075 for cached input per million tokens. Batch input and output cost $0.75 and $3.75 respectively. Access limits and any additional tool charges depend on the service configuration.

The API supports low, medium, high and xhigh thinking levels, with high as the default. Higher effort can change the time and tokens needed to complete a request; compare settings on the work you intend to run.

Access depends on the product

Hydrogen is offered through free and paid Gemini API tiers, subject to their limits. Free Gemini app users can encounter it through automatic model selection. Google’s Antigravity announcement specifies access for paid Pro and Ultra users. Those are distinct access routes, each governed by its own allowance.

Google’s model card says targeted cyber and biological testing reached alert thresholds while remaining below the corresponding full critical-capability levels. The detailed safety report is forthcoming. This page therefore makes no numerical cybersecurity benchmark comparison.

Image: Google’s official launch artwork.