Thursday, November 19, 2026

Google Unveils 'Gemini 4 Hydrogen'... Faster Coding, Greater Cost Efficiency

Input
2026-11-13 14:41:13
Updated
2026-11-13 14:41:13
Google Gemini 4 Hydrogen. Provided by Google
[Financial News] Google has unveiled Gemini 4 Hydrogen, a model designed to speed up coding and the work of AI agents. The strategy is to leave complex design to higher-tier models and use a faster model for writing and testing code, reducing the time and usage required while maintaining task success rates.
According to Google on the 13th, Hydrogen is a model focused on everyday coding and AI agent tasks that use tools across multiple steps. While Gemini 4 Carbon and Argon are used for tasks requiring deep reasoning, such as complex system design and diagnosing the causes of errors, Hydrogen is designed for writing code, running tests and verifying changes. It can be used on its own for repetitive tasks and alongside higher-tier models for complex projects.
Google is adding an 'Orchestrator' feature to Google Antigravity, its AI development tool, to switch models automatically according to each stage of a task. In a single work session, Carbon can draw up the overall plan and coordinate the work, while Hydrogen makes code changes and runs tests in parallel across multiple workspaces. This reduces the hassle of developers having to select a model at every stage and allows them to focus the use of higher-tier models on tasks that need them.
In Google's Terminal-Bench 4.0 evaluation, the Orchestrator using Carbon and Hydrogen together recorded a task success rate similar to that of Carbon used on its own. Average usage-limit consumption fell by 24%, and task completion time was cut by 19%. Google says this means more tasks can be handled within the same usage limit while maintaining the quality of the final results.
Hydrogen's standalone performance also improved over its predecessor. According to Google, with the default reasoning setting, 'High,' it completed tasks 2.5 times faster than Gemini 3.8 Flash. Its Terminal-Bench 4.1 score, which measures how well it performs practical tasks such as coding in a terminal, rose from 18.2% to 57%. On Artificial Analysis's AI Intelligence Index, it scored 51 points with the default setting, ahead of Argon's 50, and scored 54 with the highest reasoning setting.
The API costs $1.50 per million input tokens and $7.50 per million output tokens. The same rate applies to input context of up to 1 million tokens processed at a time. Four reasoning levels are available, from 'Low' to 'Very High.' Developers can adjust the reasoning level to match a task's difficulty and manage response times and token usage.
Hydrogen is available on all Google AI Studio plans, through the Gemini API and in Gemini for enterprise. Google AI Pro and Ultra subscribers can also use it in Google Antigravity.
Hydrogen will also be available in the Gemini app for general users and in Google Search. In the Gemini app, it can carry forward earlier instructions across multiple exchanges to match the format and tone of emails or documents, and handle tasks by drawing on uploaded materials and information from connected apps such as Gmail and Google Drive.
[email protected] Jang Min-kwon Reporter