Google DeepMind ships Gemini 4 Hydrogen for fast agentic work

Google DeepMind has introduced Gemini 4 Hydrogen, a Gemini 4 model built for speed rather than long-horizon reasoning, and it is already available across Google AI Studio, the Gemini API, Google Search and the Gemini app. It matters because the company is shipping it alongside an automatic model-switching feature, which changes how developers can think about picking a model at all.

Introducing Gemini 4 Hydrogen

What exactly did Google DeepMind announce?

According to Google DeepMind, Hydrogen is the fast tier of the Gemini 4 family, positioned below the frontier models Gemini 4 Carbon and Gemini 4 Argon, which the company says handle deep, long-horizon reasoning on the hardest problems. Hydrogen is aimed instead at rapid execution across agentic workflows: writing code, running tests, verifying changes, and pulling together documents and reports.

The launch post, credited to Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect at Google, says Hydrogen is coming to all tiers of Google AI Studio and the Gemini API, to Gemini for business, the Gemini app, Google Search, and to paid Google AI Pro and Ultra subscribers inside Google Antigravity.

How does it perform against other models?

Google DeepMind reports that Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, against 50 for Argon, and reaches 54 at its highest xhigh setting. The company also says Hydrogen often finishes tasks faster than models in its class.

On Terminal-Bench 4.1, Hydrogen scores 57% compared with 18.2% for Gemini 3.8 Flash, the model it is positioned to replace, while finishing tasks 2.5 times faster at its default high setting. Google DeepMind also claims it cuts fabricated claims by more than 5x.

On agentic and knowledge-work benchmarks, the post says Hydrogen ranks #5 on Agent Arena, leads its class on HLE-Diamond at 58.8%, ranks #1 overall on Vals Tax Agent Bench at 52.3%, ties for #1 on the Vals Legal Research Bench at 55.8%, and leads its class on AutomationBench-AA. It scores 53.4% on the new Multimodal Knowledge Agent benchmark and 71.6% on Agents' Last Exam v1.1.

What is the pricing and context window?

Hydrogen is priced flat across its entire 1 million-token context window at $1.50 per million input tokens and $7.50 per million output tokens, according to Google DeepMind. There is no long-context surcharge tier in the published pricing, which matters for teams that feed whole repositories or dense document sets into a single request.

The model pairs that 1 million-token input with up to 1 million output tokens. Google DeepMind says this combination is what lets it synthesize PDFs, spreadsheets, charts, audio and recorded video into cited reports and structured deliverables in one pass.

What is Orchestrator, and why does it matter?

Orchestrator is a new feature in Google Antigravity that switches between models inside the same session instead of committing to one model from start to finish. Google DeepMind says that on Terminal-Bench 4.0, Orchestrator reached near-Carbon task success rates while consuming 24% less quota on average and completing 19% faster.

The same post describes a second pattern for parallel work: Carbon acts as coordinator while Hydrogen powers worker and verifier agents that build and test changes simultaneously in isolated workspaces. For early testers running complex graphs, SQL and data-heavy enterprise workflows, Google DeepMind says partners saw Hydrogen run 2x to 2.8x faster than 3.8 Flash while using 12% to 23% fewer tokens.

What are the four thinking levels?

Hydrogen exposes four thinking levels: low, medium, high, which is the default, and a new xhigh tier. Google DeepMind describes the progression as smooth and predictable across quality, latency and cost, so a caller only waits and pays for the reasoning depth a task actually needs.

In practice that turns reasoning effort into a dial rather than a model choice. A test run or a formatting pass can sit at low or medium, while a debugging session or a dense document synthesis can be pushed to high or xhigh without changing the underlying model or the price per token.

How can you try this in Grafy?

The most direct fit is Grafy Code, the AI coding studio that can install packages and run the code it writes. Hydrogen's headline claims are about repository work — surgical edits, parallel file reads, test-and-verify loops — which is exactly the loop a coding studio exercises, so it is a reasonable place to compare a fast execution model against a slower reasoning one on your own repo.

For the multi-model side of the story, Grafy Chat puts many cloud models next to each other in one place, which is the manual version of what Orchestrator automates: ask the same prompt twice, once to a reasoning-first model and once to a speed-first one, and compare. If you want the automated version on your own material, Grafy Operator runs browsing and multi-step web tasks as an agent, and Grafy Reader is the place to test the document-synthesis claims by questioning a dense PDF and checking whether the citations hold up.

One caution from the Grafy News Desk: nearly every number in this story is a vendor-reported benchmark. Hydrogen's real value for you is measurable only on your own workload, so treat the published scores as a shortlist, not a verdict.

Frequently asked questions

Did Google DeepMind release Gemini 4 Hydrogen?

Yes. Google DeepMind announced Gemini 4 Hydrogen on 12 November 2026 as the speed-and-efficiency tier of the Gemini 4 family, sitting below the frontier models Carbon and Argon. According to the company, it is available in Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, and for paid AI Pro and Ultra subscribers in Google Antigravity.

What does Orchestrator in Google Antigravity do?

Orchestrator switches between models within a single session rather than using one model throughout. Google DeepMind reports that on Terminal-Bench 4.0 it reached near-Carbon task success rates while using 24% less quota on average and finishing 19% faster. It is designed to let a reasoning model plan while a faster model executes.

How much does Gemini 4 Hydrogen cost per million tokens?

According to Google DeepMind, Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, and that price is flat across the whole 1 million-token context window. The model accepts up to 1 million input tokens and can return up to 1 million output tokens.

Source: Google DeepMind — Introducing Gemini 4 Hydrogen

---

Source: Google DeepMind — Introducing Gemini 4 Hydrogen