On this page · 5 sections ▾
A workhorse AI model is the fast, affordable tier of a model family, tuned for the most common tasks rather than the hardest ones. Google's Gemini 4 Hydrogen is the latest example. It replaces the earlier Flash line and sits below the Argon and Carbon models, which Google positions for long, complex work. The idea is to match the size of the model to the size of the job.
Why do AI makers split models into tiers?
Running a large AI model costs more time and computing power per answer. Most requests, such as drafting an email, summarizing a document or answering a question, do not need that capacity. A smaller model can handle them quickly and cheaply.
So companies sell a ladder. In the Gemini app's model picker, as 9to5Google listed it, Hydrogen is labeled "Everyday help," Argon is for code and analysis, and Carbon is for the most complex work. A separate Flash-Lite model covers quick answers.
What are thinking levels?
Many current models can spend extra effort reasoning before they reply. Hydrogen exposes this as four settings: low, medium, high and a new xhigh. Higher settings generally mean slower, more thorough answers, and lower settings favor speed.
Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default high setting. It also says the model scores 51 on the Artificial Analysis Intelligence Index at high, against 50 for Argon, and 54 at xhigh. These are Google's own claims, so they are best read as a signal of direction rather than a final verdict.
What do tokens have to do with it?
AI services measure text in tokens, small chunks of words. Developers pay per million tokens sent in and received out. As Ars Technica reported, Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, unchanged from 3.8 Flash. Carbon and Argon are currently $2 and $10, rising to $4 and $20 next month.
That gap explains the workhorse approach. Agentic tools, which carry out multi-step tasks on their own, can burn through many tokens. Using a cheaper model for routine steps keeps bills down. Google's new Orchestrator feature in Antigravity automates this: Carbon coordinates, while Hydrogen powers worker and verifier agents that build and test changes in parallel.
Why does it matter now?
For most people, the workhorse model is the one they actually meet. Hydrogen is the default for Google's AI Mode in Search and handles the hardest questions in AI Overviews, according to 9to5Google. Free Gemini app users get it automatically when the app judges it suitable for a prompt.
