3 min read

Gemini 4 Hydrogen Does the Running While a Bigger Model Does the Thinking

Google Gemini 4 Hydrogen is a fast and cheaper model for coding and agent work, built to execute steps while heavier models handle the planning.

Creative ideas News
Screenshot of a post by kimmonismus showing a futuristic city with glowing towers in the rain, built with Gemini 4 Hydrogen from one prompt and one follow up
Image: kimmonismus via The Neuron. Story source: The Neuron, November 13, 2026

A model built to execute, not to ponder

Google has added a sprinter to the Gemini 4 family. The Neuron reports that the Gemini 4 Hydrogen launch post describes it as the fast and lower cost model for everyday coding and agent work, built to execute quickly rather than reason deeply. The heavier Gemini 4 models, Carbon and Argon, take the long and hard thinking. Hydrogen is available in the Gemini app model dropdown, in the Gemini API, and in AI Studio, and The Rundown AI listed it among the trending tools of the day.

Early testers are already showing what fast execution looks like. The Neuron shared a post by kimmonismus, who got early access to Hydrogen in Antigravity and asked it to build a futuristic city you could actually fly through. After one prompt and one follow up, the result was a rainy Three.js world with procedural towers and flying traffic.

Google already splits the job this way

This is not a theory Google is asking you to test alone. Its own coding tool, Antigravity, already splits the job between the two kinds of models. Its Orchestrator switches between Carbon and Hydrogen inside one session, and Google reports near Carbon results with about 24 percent less quota and 19 percent faster completion on its own coding tests. Read those as vendor numbers, but the direction is plain: the lab that built both models routes work this way itself.

The Neuron frames the pattern as architect and contractor. The architect decides what to build and in what order. The contractor follows the blueprint quickly. Paying the architect to hammer every nail is slow and pricey, and asking the contractor to design the building is how you get a door that opens into a wall.

Run the pattern in four moves

First, plan with your strongest model. Ask it to turn the job into numbered steps, each written to stand alone, with a done looks like check you can verify in 30 seconds, and ask it to mark any step that needs real judgment rather than execution. Second, run one step at a time on Hydrogen. Pick it in the model dropdown, paste the full plan for context, then tell it to do only that one step. Hydrogen follows tone, format, and specific rules closely, which is exactly what an execution step needs.

Third, match the thinking depth to the step. Hydrogen has four thinking levels: low, medium, high as the default, and an extra high setting that is new. You pay and wait only for the depth a task needs, so save the deepest setting for the one tricky step. Fourth, send only the failures back upstairs. If a step misses its check, hand that step and its output back to the big model for a ruling, instead of paying it to redo passing work.

Test the speed claim on your own work

One caution before you reorganize your stack around the headline number. Google measured Hydrogen at 2.5 times faster at its high default setting against an older model, Gemini 3.8 Flash. The Neuron suggests timing one of your own tasks at a lower thinking level first. Speed on a vendor chart and speed on your Tuesday workload are different measurements, and only one of them pays your bills.

The deeper point comes from the same issue of The Neuron, where Simon Willison argues that price per token no longer tells you much, because reasoning models use very different token counts. Cost per finished task is the better yardstick. For creators, that means testing Hydrogen on real steps: caption batches, research pulls, edit checklists, first drafts of show notes. Keep the planning with the model that thinks best, let the fast model run the steps, and judge the pair by finished work per dollar.

See session pricing

Sources and further reading

  1. The Neuron: Let a big model plan and Gemini 4 Hydrogen do the work (November 13, 2026)
  2. The Rundown AI: Google debuts Gemini 4 Hydrogen (November 13, 2026)

Sources were checked for this article. Tool features, prices and availability can change. Product claims are attributed to their publishers.

Keep exploring.