Gemini 4 Hydrogen: Google's Fast AI for Coding

Flat illustration of a rocket-shaped AI chip speeding through a teal and yellow sky, with the words "HYDROGEN. FAST BY DESIGN."

Google has a new workhorse model, and it is built for speed. On 12 November 2026, the company introduced Gemini 4 Hydrogen, a model designed for everyday coding and agentic workflows — the kind of jobs where an AI writes code, runs a test, reads the failure, and tries again. It replaces Gemini 3.8 Flash across Google's products, and it is rolling out to free users as well as developers.

If you use Google Search or the Gemini app, you may already be using it. Google says Hydrogen is now powering Search globally, and it appears in the Gemini app's model picker. For developers, it is available through the Gemini API and Google AI Studio on every tier, free and paid.

What Gemini 4 Hydrogen actually is

Hydrogen is the fast member of the Gemini 4 family. Google's heavier models, Carbon and Argon, handle deep, long-horizon reasoning. Hydrogen is the execution engine: it does the unglamorous half of agentic work, writing the change, running the test, and checking the result.

It accepts text, images, audio, and video, and it produces text. Its context window holds up to one million tokens, counting input and output together. The model is adapted from Carbon, and its knowledge cutoff is March 2026.

One feature stands out: adjustable thinking. Hydrogen offers four thinking levels — low, medium, high, and a new xhigh setting for harder problems. The default is high, a step up from the medium default of Gemini 3.8 Flash. Turn the dial up and the model thinks longer; turn it down and it answers faster.

How fast is it, really?

Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default high reasoning setting. That is a company claim, measured on Google's own tests, and it is worth treating as a claim rather than an independent result.

The independent evaluator Artificial Analysis tested the model on its Intelligence Index. At the default high setting, Hydrogen scored 51, just ahead of Argon's 50. At the xhigh setting it reached 54 — but the same evaluator found that moving above the default more than doubled the estimated task time. Speed and smarts trade off against each other, even inside one model.

Google's own model card adds more detail. On SWE-sweep, a coding benchmark, Hydrogen scored 13.5%, compared with 3.1% for Claude Sonnet 5.5 and 5.8% for GPT-6.1 Terra. On Terminal-Bench 4.1, it scored 57.0%, below Sonnet 5.5's 63.9%.

What it costs

Pricing is simple and flat: $1.50 per million input tokens and $7.50 per million output tokens, across the entire one-million-token context window. There is no cheaper rate for short prompts and no premium for long ones.

For everyone else, it is free in Search and the Gemini app. Paid Google AI Pro and Ultra subscribers also get Hydrogen in Antigravity, Google's coding environment, where a new feature called Orchestrator switches between Carbon and Hydrogen inside one session — Carbon plans, Hydrogen executes.

The catch worth knowing

Hydrogen is not a do-everything model, and Google's own documentation says so. Three limits matter if you are building with it.

First, it is a short-term generally available release, subject to a minimum 45-day retirement notice. That means Google could retire it with as little as 45 days' warning. If you build a product on it, plan for a switch.

Second, the model page lists RAG Engine, tuning, and the Gemini Live API as not supported. If your app needs retrieval-augmented generation or fine-tuning, Hydrogen is not the right pick today.

Third, the full safety and security report is still to come. Google says it will be available in the coming days. Until then, the safety claims rest on the announcement alone.

Who should use it?

Hydrogen fits developers building coding assistants and tools that carry out tasks with other software: repository coding, debugging, interactive web development, and multimodal jobs. It also fits anyone who wants a faster Gemini in Search or the Gemini app without paying.

Key takeaways

  • Gemini 4 Hydrogen launched on 12 November 2026, replacing Gemini 3.8 Flash across Google Search, the Gemini app, and developer tools.
  • It is built for fast coding and agentic work, with four thinking levels from low to xhigh.
  • It costs $1.50 per million input tokens and $7.50 per million output tokens, flat across a one-million-token window.
  • It scored 51 on the Artificial Analysis Intelligence Index at its default setting, and 54 at xhigh — with much longer task times at xhigh.
  • It is a short-term release with a 45-day retirement notice, and it does not support RAG Engine, tuning, or the Live API.

Conclusion

Gemini 4 Hydrogen is Google's answer to a practical problem: most agentic work is not one brilliant insight, it is hundreds of small steps done quickly. A model that executes fast, at a flat price, and sits inside Search and the Gemini app for free is a useful thing. The 45-day retirement notice and the missing features are real limits, but for everyday coding and agent work, Hydrogen is now the default Gemini — and most users will meet it without ever choosing it.

FAQ

Is Gemini 4 Hydrogen free?

Yes, for everyday use. It is free in Google Search and the Gemini app. Developers pay $1.50 per million input tokens and $7.50 per million output tokens through the Gemini API and Google AI Studio, which also have free tiers for prototyping.

What does Gemini 4 Hydrogen replace?

It replaces Gemini 3.8 Flash as Google's fast model tier. Google says Hydrogen finishes tasks 2.5 times faster than Flash at its default setting, while scoring higher on independent benchmarks.

Can I use Gemini 4 Hydrogen in Antigravity?

Yes, if you are a paid Google AI Pro or Ultra subscriber. Antigravity also gains Orchestrator, which switches between Carbon and Hydrogen within a session.

Researched with AI assistance and reviewed for accuracy.

Sources

Follow on WhatsApp

Follow Daily Tech & AI on WhatsApp for the latest tech and AI news, every day: Daily Tech & AI on WhatsApp

Comments