Google puts Gemini 4 Hydrogen in front of everyone: teach people to check it, not just use it

Google released Gemini 4 Hydrogen on 12 November 2026. In a post by Koray Kavukcuoglu, SVP at Google DeepMind and Google's Chief AI Architect, Google describes it as a fast, cost-efficient model for agentic workflows and everyday coding. It is priced at $1.50 per million input tokens and $7.50 per million output tokens across a one-million-token context window. Google says it finishes tasks 2.5 times faster than Gemini 3.8 Flash, scores 57% against 18.2% on the Terminal-Bench 4.1 coding test, and produces more than five times fewer fabricated claims. These are Google's own figures and have not been independently verified.

Rollout began on 12 November. Developers get Hydrogen through the Gemini API, AI Studio and Android Studio. Businesses get it through Gemini for business, Google Antigravity and, for eligible subscribers, Gemini in Google Sheets. Consumers see it in the Gemini app model drop-down, and free users are served it through automatic model selection when appropriate. Google also lists Google Search for all users. Its post makes no mention of schools, teachers or children.

The change that matters beyond the benchmarks is where the model sits. A model that is cheaper to run and reaches free users and Search is one that pupils, parents, teachers and employees meet without choosing it. Most people using Search or a free app are not picking a model at all. The answer simply appears, at a kitchen table or on a deadline, and the habit of checking it has to travel with the person, because the tool will keep changing underneath them.

Google reports that Hydrogen reduces misaligned actions in agentic and coding tests twelvefold and records a 96.7% honesty rate on its internal agent evaluation. That direction is welcome. But fewer errors is not no errors, and the human risk with a faster, cheaper model is that people check it less. Evolution, not revolution, is the right frame: the tools improve steadily, and the thinking people do around them has to improve at the same pace. Teachers and workers are already ahead of policy on this, and paralysis is a bigger risk than adoption.

For school leaders, tell staff and pupils plainly that the default AI behind Search and the Gemini app can change without notice, and build a checking step into any task where an AI answer is used: ask pupils to pick one claim and verify it against a source. Where staff use AI to run tasks or write code, name the person who reviews what it did. Business leaders should ask suppliers which model sits underneath their tools and be told when it changes. Parents can ask a child to show how they checked an answer, not only what the answer was.

Dan Fitzpatrick works with schools and organisations on using AI well while keeping people in charge of the thinking, and writes and speaks regularly on AI literacy, learning and work.

Dan Fitzpatrick is available for interview and comment on this story. He is available today for live or pre-recorded radio, TV, podcast and print interviews, and has a broadcast-quality remote setup for live or pre-recorded down-the-line interviews. Contact: dan@theaieducator.io