Google Gemini 4 Hydrogen Guide

Gemini 4 Hydrogen is Google’s new fast model for everyday coding and agent work, released on November 12, 2026. Google calls it its “daily driver” for agentic workflows.

Carbon and Argon think deeply. Hydrogen gets everyday work done fast.

Gemini 4 Carbon

Google’s strongest model, for the hardest problems

Gemini 4 Argon

Deep reasoning for complex coding and knowledge work

Gemini 4 Hydrogen

Fast, efficient execution for everyday coding and agent tasks

Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default setting, slightly above Argon’s 50, and 54 at its highest setting.

• Finishes tasks about 2.5 times faster at its default setting, per Google
• 57% on Terminal-Bench 4.1, versus 18.2% for 3.8 Flash
• More than five times fewer made-up claims
• Follows instructions more closely and calls tools more reliably

Thinking levels

Low, medium, high (the default), and a new xhigh

Context window

Up to 1 million input tokens, with up to 1 million output tokens

Use low for quick answers and xhigh when quality matters more than speed.

Gemini app

In the model menu; free users get it when automatic model selection picks it

Google Search

Rolling out to all users worldwide

Google Antigravity

Google AI Pro and Ultra subscribers

Developers

Gemini API, Google AI Studio, and Android Studio

Businesses

Gemini for business, and Gemini in Google Sheets for eligible plans and Pro and Ultra subscribers

Input

$1.50 per million tokens

Output

$7.50 per million tokens

Google says the price is the same across the full 1 million token context.

• 58.8% on HLE-Diamond, leading its class
• First on Vals Tax Agent Bench at 52.3%
• Tied first on Vals Legal Research Bench at 55.8%
• Early partners saw 2 to 2.8 times faster runs and 12% to 23% fewer tokens than 3.8 Flash

These are Google’s own figures; test it on your own work.

• In Antigravity, Orchestrator can switch between Carbon and Hydrogen in one session
• Carbon can plan and coordinate while Hydrogen runs the worker and checker agents
• Google says this got close to Carbon’s results with about 24% less quota and 19% faster runs on one benchmark

• Google reports up to 81% less agentic overreach
• Up to twice as unlikely to bypass permission limits when a task is blocked
• More than 2.3 times more resilient to indirect prompt injection on one benchmark

Keep approvals on for risky actions anyway; no model is fully immune to manipulation.

• Use Hydrogen for quick edits, data tasks, and everyday agents
• Switch to Carbon or Argon for long, hard problems
• Raise the thinking level only when results need it
• Watch token use on long agent runs