MONDAY, NOVEMBER 16, 2026 Tech news, minus the noise.
N News Cast
LATEST
Home  /  AI
AI

Gemini 4 Hydrogen Is Google's Fast Model for Coding Agents, and the 2.5x Speed Claim Needs a Closer Look

Google shipped Gemini 4 Hydrogen on November 12 as the fast tier of its Gemini 4 family, priced flat at $1.50 per million input tokens across a 1 million token context window.

Gemini 4 Hydrogen Is Google's Fast Model for Coding Agents, and the 2.5x Speed Claim Needs a Closer Look
Source: Sergei Magel / Heinz Nixdorf MuseumsForum (CC BY-SA 4.0) via Wikimedia Commons

Gemini 4 Hydrogen arrives as Google's fast tier

Gemini 4 Hydrogen is Google's new fast model for coding and agentic work, and it went generally available on November 12, 2026. It is the speed tier of the Gemini 4 family, built for the jobs where a model runs dozens or hundreds of times a day: writing code, calling tools, triaging tasks, and driving agents that work in loops. Google is selling it flat at $1.50 per million input tokens and $7.50 per million output tokens across the full 1 million token context window, with no price jump for long prompts.

That flat pricing is the first thing worth noticing. Gemini 4 Carbon, the frontier model in the same family, costs $2 per million input tokens until November 30 and then doubles to $4. Gemini 4 Argon follows the same path. Hydrogen stays at $1.50. If you run agents at volume, the model that never gets more expensive as your context grows is the one your bill actually depends on.

What Gemini 4 Hydrogen actually does

Hydrogen comes with four thinking levels: low, medium, high, and a new xhigh setting. At its default high level it scores 51 on the Artificial Analysis Intelligence Index, just ahead of Argon's 50. At xhigh it reaches 54. Google says tasks finish 2.5 times faster than with Gemini 3.8 Flash, the model Hydrogen replaces in the fast slot. It is also the first model in that slot sold under a codename instead of the Flash label, which tells you Google now treats its fast tier as a real product line rather than a cut-down version of the flagship.

The model is adapted from Gemini 4 Carbon, according to Google's own model card, so it inherits the frontier model's knowledge with a March 2026 cutoff. It is rolling out everywhere at once: the Gemini API and AI Studio on free and paid tiers, the Gemini app, Google Search, Gemini for business, and Google Antigravity, where it now powers the agent in Managed Agents by default. Android Studio gets it too. That breadth matters, because a fast model only changes your work if it shows up in the tools you already use.

The 2.5x speed claim, checked against independent testing

Google's headline number deserves scrutiny. Independent testing on the same harness Google cites found Hydrogen completing index tasks in 13.5 minutes against 14.8 minutes for Gemini 3.8 Flash. That is faster, but it is not 2.5 times faster. The gap between vendor benchmarks and independent harnesses is normal in this industry, and it is exactly why you should treat launch-day numbers as marketing until someone else reproduces them. Hydrogen's own index score of 54 at xhigh is real and independently measured, and that is the number to quote.

There is another wrinkle. At xhigh, Hydrogen takes 37.8 seconds to produce its first answer token, against 16.7 seconds at the default high setting and 17.3 for Gemini 3.8 Flash. If your agent waits on first tokens, the highest effort level can feel slower even while it finishes the whole task sooner. The practical move is to leave the model on high for interactive work and reserve xhigh for background jobs where total time matters more than first response.

What this means for developers building agents

Hydrogen lands in a week when the model market split in two directions. Anthropic's Claude Fable 5.2 took the top of the same index at 65, and Meta's Muse Ember undercut everyone on price for repetitive agent tasks. Google's answer is a model that is cheap, fast, and everywhere. For teams building agents, that combination is often worth more than a few extra benchmark points.

The pricing calendar adds urgency to the choice. Carbon and Argon double in price on December 1, while Hydrogen holds steady. If your agent pipeline runs on Carbon today, November is the month to test whether Hydrogen does the same job for less. The model card gives you 45 days of retirement notice at minimum, so there is time to evaluate, but not much time to dawdle.

Google's Gemini 4 Carbon remains the stronger model for the hardest reasoning work, and the Gemini Decisions API covers cheap classification calls. Hydrogen fills the middle: smart enough for real coding, cheap enough to run all day, and fast enough that your agent does not keep users waiting. That middle is where most agent workloads actually live.

AS
News Cast Editorial Team

We cover AI, gadgets, startups and security — original reporting and explainers, written by humans, for humans. Got a tip? Reach us via the contact page.