Gemini Decision 1
Multimodal decision model returns calibrated probabilities for binary, categorical, and rubric questions without generating output tokens.
Gemini 4 Hydrogen
aistudio.google.com
Generally available coding and agent model adds xhigh thinking and access through free and paid API tiers.
Visit website
Pricing
Freemium
Available on
Google AI Studio, API, Google Cloud
Category
Llm Models
What is Gemini 4 Hydrogen?
Gemini 4 Hydrogen is a Google model for developers building coding assistants, web applications and agents. Its intended uses include coding, iterative development and multimodal document workflows. Developers access it through Google AI Studio or supported APIs using the model identifier gemini-4-hydrogen.
The model can analyze documents and audiovisual material, while reasoning controls let teams balance response time against token consumption. Its large context window accommodates input, reasoning and generated output together, making context allocation an important integration decision.
Key features
Four thinking levels
Choose low, medium, high or xhigh reasoning. High is the default; deeper reasoning can increase latency and token use.
Unified context window
Input, thinking and output share 1,048,576 tokens. A synchronous segment can generate up to 65,536 output tokens; longer responses require continuation or background execution and remain bounded by the shared window.
Multimodal analysis
Analyze tables, charts, PDFs, images, audio and video alongside textual instructions.
Tools and structured responses
The Cloud offering supports function calling, structured output, code execution and grounding tools for application and agent workflows.
Office document inputs
Cloud APIs accept supported Microsoft Office and OpenDocument formats in addition to PDF and plain text.
Who is it for?
Developers building coding assistants
Evaluate the model for proposed code changes, debugging support and iterative application development.
Teams processing mixed documents
Prototype a workflow that extracts and organizes information from documents, charts and visual material.
Engineers building tool-using agents
Connect selected application functions and assess whether generated tool calls meet the task requirements.
How to use Gemini 4 Hydrogen
- Open Google AI Studio and select Gemini 4 Hydrogen for an initial evaluation.
- For integration, follow the developer guide to configure API access and a Python, JavaScript or REST client.
- Select gemini-4-hydrogen and provide your task with any supported input material.
- Start with the default high thinking level, then evaluate other levels against your latency and quality requirements.
- Monitor token consumption and validate generated responses or tool calls before production use.
Gemini 4 Hydrogen pricing
Freemium
The Gemini API offers free and paid access. Standard paid inference is billed by token usage, including thinking tokens in output charges. Grounding and other service tiers have separate pricing.
Gemini API Standard paid tier
USD 1.50 per million input tokens; USD 7.50 per million output tokens
Usage-based token billing; no monthly subscription interval
Rates apply to Gemini 4 Hydrogen Standard inference. Output pricing includes thinking tokens. Grounding can incur additional charges.
What to know before you use it
- The Cloud model ignores custom temperature, top_p and top_k values. Explicit context caching, tuning and Gemini Live API access are unsupported.
- Free Standard API access excludes Google Search and Google Maps grounding. Free-tier content is eligible to improve Google products; paid-tier content is not used for that purpose.
- The model can hallucinate, respond slowly or time out. Its documented knowledge cutoff is March 2026, with some domains potentially less current.
Frequently asked questions
Does multimodal support include image or video generation?
Hydrogen accepts image, audio and video inputs for analysis and produces text output. These input capabilities do not provide image or video generation.
Which thinking level should I start with?
High is the default. Evaluate low, medium or xhigh against your task requirements, since reasoning depth affects latency and token consumption.
Does a long context window mean unlimited output?
No. Input, thinking and output share a finite window. Synchronous output is limited to 65,536 tokens per segment, and longer generation requires continuation or background execution.