Robots Atlas>ROBOTS ATLAS
Gemini 4 Hydrogen

Gemini 4 Hydrogen

4 Hydrogen · Family: Gemini
Google model (Gemini 4 family) for fast agentic workflows; 1M-token context, available via the Gemini API and Google AI Studio.
✓ Active✓ Public accessLLMMultimodalReasoning modelTool-using model📁 Gemini
Context window
1M
tokens
Max output
1,000,000
tokens
Release date
12 November 2026
Access:APIHostedDeployment:☁ Cloud

Overview

Gemini 4 Hydrogen is an AI model developed by Google (Google DeepMind), introduced on 12 November 2026 as a Gemini 4 family variant built for rapid execution across diverse agentic workflows. Unlike the frontier Carbon and Argon models aimed at deep, long-horizon reasoning, Hydrogen serves as the execution model in agentic pipelines and can act as the execution component in multi-model workflows.

The model provides a native 1M-token input context window and generation of up to 1M output tokens, with pricing held flat across the entire context window. It exposes four thinking levels (low, medium, high as default, and a new xhigh) and native parallel batching of file-read and search calls. Model weights are not publicly available (closed model).

Hydrogen is rolling out globally through the Gemini API and Google AI Studio (all tiers), Android Studio, Gemini for business, Google Antigravity (Google AI Pro and Ultra subscribers), Gemini in Google Sheets (Google AI Pro and Ultra), the Gemini app, and AI Mode in Google Search for all users. API pricing is $1.50 per 1M input tokens and $7.50 per 1M output tokens (thinking tokens included), with cached input reads at $0.075.

Classification
LLMMultimodalReasoning modelTool-using model
Family: Gemini
Access & deployment
APIHosted
Cloud
Weights: Closed
Key parameters
📏 Context: 1M
✓ Tools
📥 Input: text, image, audio, video…

Technical specification

Context window
1M
tokens
Max output tokens
1,000,000
tokens per response
Features:✓ Tool use
Modalities
⬇ Input
textimageaudiovideodocumentsstructured_data
⬆ Output
textcodestructured_data

Capabilities and applications

Native model capabilities
Reasoning
The model's ability to reason logically and solve complex problems.
Category: reasoning
Multi-step reasoning
Carrying out multi-step chains of reasoning across long, complex tasks.
Category: reasoning
Coding
Generating, analysing and modifying code in many programming languages. Covers writing functions, debugging, refactoring, code review, and creating tests. Measured by benchmarks such as HumanEval and SWE-bench.
Category: coding
Agentic coding
Multi-hour, multi-step programming tasks performed autonomously by the model: cloning a repository, running tests, iterating on fixes, integrating with CLI tools. Characteristic of Codex variants (GPT-5.1-Codex-Mini, Codex-Max).
Category: coding
Agentic capability
The model's ability to autonomously plan and execute multi-step tasks by sequentially using tools, maintaining context, and adapting to intermediate results.
Category: planning
Multi-step project execution
The ability to autonomously drive multi-hour, multi-step projects: decomposing the task, planning the sequence of actions, iteratively delivering results, and adjusting based on feedback. Key for enterprise and knowledge-work agents.
Category: planning
Tool use
The model's ability to call external functions, APIs and tools during a conversation: calculator, search engine, code editor, database. The model decides when and how to use a tool and interprets its result.
Category: planning
Web browsing
Ability of the model to autonomously search and browse web pages to retrieve up-to-date information.
Category: other
Long context
Support for large context windows — tens to hundreds of thousands (or millions) of input tokens. Enables analysis of entire codebases, long documents, and many parallel conversations without losing earlier information. GPT-5.1 supports 400,000 tokens.
Category: language
Multimodal understanding
Category: multimodal
Image understanding
Analysing and interpreting the content of images.
Category: vision

Benchmark results

7 benchmarks
Artificial Analysis Intelligence Index
51 (high) / 54 (xhigh)
📅 12 Nov 2026📄 Oficjalna strona (blog Google)
Score of 51 at the high thinking level (vs 50 for Gemini 4 Argon) and 54 at the new xhigh level.
Terminal-Bench 4.1
57%%
📅 12 Nov 2026📄 Oficjalna strona (blog Google)
57% versus 18.2% for Gemini 3.8 Flash.
HLE-Diamond
58.8%%
📅 12 Nov 2026📄 Oficjalna strona (blog Google)
Leading result in its model class.
Vals Tax Agent Bench
52.3%%
📅 12 Nov 2026📄 Oficjalna strona (blog Google)
Ranked #1 overall.
Vals Legal Research Bench
55.8%%
📅 12 Nov 2026📄 Oficjalna strona (blog Google)
Ties for #1.
Agents' Last Exam v1.1
71.6%%
📅 12 Nov 2026📄 Oficjalna strona (blog Google)
Agentic capability evaluation.
Multimodal Knowledge Agent (MKA)
53.4%%
📅 12 Nov 2026📄 Oficjalna strona (blog Google)
Multimodal knowledge agent benchmark.

Pricing

Technical architecture

Deployment and security

🔒 Security / Enterprise
✓ Verified enterprise information
Updated: 18 Nov 2026↗ Security documentation