Gemini 4 Hydrogen: Google's faster, cheaper AI model for agents

Google released Gemini 4 Hydrogen on Thursday, 12 November 2026, a faster and cheaper member of its Gemini 4 family of artificial intelligence (AI) models, built for "agentic" work in which AI carries out multi-step tasks such as writing and testing code. According to Google's announcement, the Gemini Hydrogen model is available to developers through Google AI Studio and the Gemini API, and is rolling out in the Gemini app and Google Search. Reported API pricing is USD 1.50 per million input tokens and USD 7.50 per million output tokens.

An AI "agent" is a model that does not just answer a question but takes a series of actions: opening files, running code, checking results and trying again. These tasks need many quick model calls, so cost and speed matter as much as raw intelligence. Hydrogen is Google's attempt to fill that role, while its larger models plan and check the work. In Antigravity, for example, Google says the bigger Carbon model can coordinate a task while Hydrogen runs "worker" and "verifier" agents that do and check the individual steps.

The "thinking levels" let developers choose how long the model reasons before answering. A higher level usually gives better answers but costs more time and money. Google's own benchmark figures show Hydrogen improving as the thinking level rises, but these are company-reported numbers, not independent tests, and readers should treat them as Google's claims.

The claim about reducing "overreach" deserves attention. One risk with AI agents is that they take actions the user did not ask for, such as editing the wrong files. A model that sticks more closely to its instructions is safer to hand real work. Google puts the improvement at up to 81%, a figure that has not been independently checked.

For Indian developers and start-ups, cheaper agent models lower the cost of building AI products, from coding assistants to customer-service bots. No India-specific pricing or availability details were announced in the reports reviewed, and access in the Gemini app appeared to vary by subscription plan.

Peepals calculation: At the reported prices, output tokens cost five times as much as input tokens (7.50 ÷ 1.50 = 5). A job that reads 10 million tokens and writes 2 million would cost about USD 30 (10 × 1.50 + 2 × 7.50 = 15 + 15). Cached input, at USD 0.075 per million, is 95% cheaper than fresh input (0.075 ÷ 1.50 = 0.05). And Google's Terminal-Bench 4.1 score of 57% is a little over three times the 18.2% of Gemini 3.8 Flash (57 ÷ 18.2 ≈ 3.1). Our calculation from the reported figures.

Short-term: Developers can start building with Hydrogen through the Gemini API and AI Studio, and consumers may see it in the Gemini app and Search depending on their plan.

Long-term: Cheaper, faster agent models push the AI industry towards systems in which a large model plans and several small models carry out the steps, lowering the cost of automating software and office work.

Who is affected: Software developers and start-ups building AI tools, businesses using Gemini for work, Gemini app users, and rival AI companies competing on price and speed.

Questions and Answers