Google has released Gemini 4 Hydrogen, the speed-focused member of its Gemini 4 family. Where the flagship models chase maximum capability, Hydrogen is built to be quick and cheap enough to put everywhere — and Google is rolling it out across both developer and consumer surfaces, including to free users.

The pricing is deliberately simple: a flat $1.50 per million input tokens across the model's full one-million-token context window, with no surcharge for long inputs. On capability, Google cites a 57% score on Terminal-Bench 4.1, the benchmark that measures how well models complete real tasks in a terminal.

The release continues the pattern of this model generation: the frontier labs no longer ship one model at a time, but families — a heavy model for hard problems, a fast model for everything else, and specialised variants in between. Hydrogen is the one most people will actually meet.

The background

Every major lab now ships a family rather than a model. The flagship exists to prove what is possible and to win the hardest evaluations; the fast tier exists to be used — cheap enough to embed in products, quick enough to answer in real time, and good enough that users rarely notice the trade. Most real-world AI traffic ends up on the fast tier, which makes it, commercially, the most important model in the family.

Hydrogen's design choices all point the same way. A million-token context window at a flat price removes the penalty for long inputs — whole document collections, codebases and conversation histories can be handed to the model without a pricing surprise. The Terminal-Bench score Google quotes, 57%, is a deliberate choice of evidence: not exam questions, but completing real tasks in a terminal.

Why it matters

Reaching free users is the strategic core of this release. The free tier is where habits form and where the next generation of developers learns which model to reach for first. A fast model that free users can actually use, across consumer and developer surfaces at once, is a distribution weapon as much as a technical one.

The flat $1.50 pricing also pressures the market. When a lab of Google's scale prices a million tokens of fast, long-context inference this simply, rivals' complicated tiered pricing starts to look like a tax on their own customers — and developers, who feel every pricing wrinkle, notice.

The wider story

There is an energy angle hidden in the fast-model race. The cheaper and quicker a model is to run, the more the industry can afford to run it everywhere — in every search box, document and app. The frontier gets the headlines, but the fast tier is where AI becomes ambient, and where the real volume of the AI economy will live.

What happens next

Watch how quickly Hydrogen becomes the default across Google's own products, and whether competitors answer with their own flat-priced fast tiers. The flagship race decides prestige; the fast-model race decides who everyone actually uses.

Common questions

What is Gemini 4 Hydrogen?

Gemini 4 Hydrogen is the fast member of Google's Gemini 4 family, released on November 12. It is built to be quick and cheap enough to use everywhere, and it is rolling out across developer and consumer surfaces, including to free users.

How much does Gemini 4 Hydrogen cost?

Pricing is a flat $1.50 per million input tokens across the model's full one-million-token context window, with no surcharge for long inputs. Google cites a score of 57% on Terminal-Bench 4.1 for the model.