Google launches Gemini 4 Hydrogen for coding and agent tasks

A new tool for complex workflows

Google has officially released Gemini 4 Hydrogen, a specialized artificial intelligence model. This new iteration is designed to handle specific types of heavy lifting, such as coding, processing large documents, and executing multi-step agent-based tasks. These tasks require the AI to plan several moves in a row rather than just answering a single prompt.

The model represents a shift toward more capable reasoning for enterprise applications. By focusing on these complex workflows, Google aims to provide a tool that can act more like an autonomous agent. This means the AI can follow a series of instructions to complete a project from start to finish with fewer human interventions.

While it is a significant step forward in capability, it is not the only model in Google's lineup. It is positioned as a high-performance option for users who need more than what standard, faster models can provide. The launch marks a move toward specialized models that trade off some speed for deeper intelligence.

Performance gains in coding

One of the most notable improvements for Gemini 4 Hydrogen is seen in terminal-based coding tasks. According to evaluations from Google, the model completed 57% of tasks on the Terminal-Bench 4.1. This is a substantial jump from Gemini 3.8 Flash, which only managed to complete 18.2% of those same tasks.

Independent testing by Artificial Analysis also highlighted these improvements. The new model received an Intelligence Index score of 51. In comparison, Gemini 3.8 Flash received a score of 31 when tested at high reasoning levels. These numbers suggest that Hydrogen is significantly more capable of handling difficult logic problems.

Despite these gains, it is not the top performer in every category. In the same Google comparison, Claude Sonnet 5.5 outperformed Hydrogen with a score of 63.9%. This indicates that while Hydrogen is a major upgrade over previous Google models, other competitors still hold the lead in certain coding benchmarks.

Speed and efficiency metrics

Speed is a primary selling point for the new model. Google reports that Gemini 4 Hydrogen finishes work 2.5 times quicker than Gemini 3.8 Flash when using its default reasoning setting. This speed increase is particularly relevant for businesses trying to scale up automated processes that require high-quality output.

Early access testing by the company Box provided real-world evidence of these efficiency gains. Their AI agent saw its accuracy rise from 74% with the older Flash model to 79% with Hydrogen. Box CEO Aaron Levie noted that the new model finished typical tasks in less than half the time. This was achieved while using around 12% fewer tokens.

Tokens are the basic units of text that AI models process to understand and generate language. By using fewer tokens to achieve higher accuracy, the model demonstrates a more efficient way of processing information. This allows for more complex work to be done without necessarily increasing the amount of data the model has to read.

New pricing and access tiers

The increased performance of Gemini 4 Hydrogen comes with a higher price tag for developers. The standard API price is set at $1.50 for every million uncached input tokens and $7.50 for every million output tokens. These costs apply across the model's entire context window, which is the amount of information the model can consider at one time.

For users who need to manage costs, Google is offering several different plans. Batch, Flex, and Contributor options are available at half the standard rates. However, these discounted options come with different processing conditions that users must consider before switching from standard pricing.

The Contributor plan is unique because it is invite-only. If a user chooses this plan, they allow Google to use their request data to train models and improve their products. This creates a trade-off between lower costs and the privacy of the data being fed into the system. It is a common model for companies looking to save money while contributing to the ecosystem.

What comes next for agents

The launch of Gemini 4 Hydrogen signals a growing trend toward agentic AI. These are systems that do not just chat with users but actually perform work. By improving accuracy and speed in terminal environments, Google is making it easier for developers to build tools that can write, test, and deploy code autonomously.

For companies already using Google's AI, the transition to Hydrogen will likely involve a cost-benefit analysis. They will need to determine if the accuracy boost and the 2.5x speed increase justify the higher API costs compared to the Flash model. Many will likely start with the Batch or Flex plans to test these capabilities at a lower price point.

As the market for AI agents matures, the focus will shift from simple chat interfaces to reliable task completion. Hydrogen is a clear attempt by Google to capture the high-end of this market. Future updates will likely focus on further refining these multi-step tasks to reduce the need for human oversight in complex coding environments.