Google Puts Hydrogen, a Model 2.5x Faster Than Gemini 3.8 Flash, Into Search
Google has updated Search so it handles long, multi-part questions and queries with typos better. The work is done by Gemini 4 Hydrogen, a fast model that Google says runs 2.5x faster than the previous Gemini 3.8 Flash. At the high reasoning level, it also outscored the more powerful Gemini 4 Argon on the Artificial Analysis index. The takeaway for developers: heavy models are not always needed for agents.
AI-processed from TechCrunch; edited by Hamidun News
You type a long question with several conditions into Google, and Search latches onto a single word and returns an answer to something else. Or you make a typo, and the system misreads you. Google says these cases are now handled better. The update arrives alongside a new model, Gemini 4 Hydrogen. According to the company, it completes tasks 2.5x faster than Gemini 3.8 Flash, Google's previous fast model. The update is rolling out to all Search users worldwide, and developers get the same model in their tools.
What changed in Search
Google calls Hydrogen "the biggest model improvement in Search ever." In practice, this means you can ask a question the way it comes to mind: long, in several parts, with very specific conditions. For example: "Which laptop suits video editing, weighs under 1.5 kilograms, and doesn't cost as much as a used car?" Previously, such queries had to be split up by hand.
Hydrogen follows instructions better and calls the right tools more accurately, so it answers each part of the question separately. It also handles hard-to-read typos better.
There is a less visible improvement too. Search uses a technique called fan-out: a user's query is split into several sub-queries, and each one is searched separately. According to Google, Hydrogen's stronger reasoning makes this splitting more effective, so the results more often include sites a person might otherwise have missed. In AI Mode, there will be more images, structured answers, comparison tables, and built-in mini tools. Answers have also improved in the languages AI Mode supports.
Why a fast model alongside a heavy one
Within the Gemini 4 lineup, Hydrogen has a clear role. The flagship Gemini 4 Carbon and Gemini 4 Argon are for deep reasoning on hard tasks. Hydrogen is built to run agentic workflows quickly, that is, tasks where the AI goes through several steps in a row. Google's logic: the most powerful model is not always needed; sometimes a fast, cheap, and sufficiently smart one is enough.
The figures the company cites:
- Hydrogen completes tasks 2.5x faster than Gemini 3.8 Flash;
- on the Artificial Analysis Intelligence Index, a general intelligence benchmark for models, at the high reasoning level (the default), it outperforms the more powerful Gemini 4 Argon;
- on multi-step tasks such as deep research, data analysis, and navigating a large codebase, the model ranks 5th in Agent Arena.
There are four reasoning levels: low, medium, high, and the new xhigh. According to Google, all of them are designed so tasks get solved faster and with fewer output tokens. For those paying for API access, this means less output text for the same tasks. The announcement gives no concrete prices, so you will have to calculate the savings yourself.
What developers should do with this
Hydrogen can work on its own for everyday tasks: writing code, running tests, checking edits. It can also be paired with Carbon and Argon. In that setup, the fast model takes the routine work, and the heavy models are called only for complex decision points. Many people building agents already use this pattern, and now Google has ready-made models for it.
Availability is broad: Hydrogen comes to all Google AI Studio plans, the Gemini API, and the Gemini app. AI Pro and Ultra subscribers will get it in the Antigravity development environment.
Some caveats are worth keeping in mind. All comparisons come from Google's own announcement, and the material contains no independent tests. The Artificial Analysis index measures general intelligence and does not guarantee that Hydrogen will beat Argon on your specific task. The fifth place in Agent Arena also shows the model is strong, but not leading.
Practical advice: if you are building an agent, run your typical scenarios on Hydrogen in high and xhigh modes. Compare speed, quality, and token usage against heavier models. If the quality difference is small and the answer arrives many times faster, you can reserve the expensive model for rare, complex steps only.
- Nov 21, 2026AcpGram Lets You Control Claude Code, Codex, and Gemini CLI Directly from Telegram
- Nov 21, 2026xAI Releases Grok Bot 1.0: One AI Bot for the Whole Team in Slack, Teams, and Video Calls
- Nov 21, 2026Back-of-the-envelope math: splitting tasks across models cuts a contact center's bill from 207,000 ₽ to 23,000 ₽
- Nov 21, 2026A $2-a-Month Script Checks Whether ChatGPT Cites the Author's Site; llms.txt Doesn't Help
Key Questions Answered
What changed in Search?
Google calls Hydrogen "the biggest model improvement in Search ever." In practice, this means you can ask a question the way it comes to mind: long, in several parts, with very specific conditions. For example: "Which laptop suits video editing, weighs under 1.5 kilograms, and doesn't cost as much as a used car?" Previously, such queries had to be split up by hand. Hydrogen follows instructions better and calls the right tools more accurately, so it answers each part of the question separately.
Why a fast model alongside a heavy one?
Within the Gemini 4 lineup, Hydrogen has a clear role. The flagship Gemini 4 Carbon and Gemini 4 Argon are for deep reasoning on hard tasks. Hydrogen is built to run agentic workflows quickly, that is, tasks where the AI goes through several steps in a row. Google's logic: the most powerful model is not always needed; sometimes a fast, cheap, and sufficiently smart one is enough.
What developers should do with this?
Hydrogen can work on its own for everyday tasks: writing code, running tests, checking edits. It can also be paired with Carbon and Argon. In that setup, the fast model takes the routine work, and the heavy models are called only for complex decision points. Many people building agents already use this pattern, and now Google has ready-made models for it. Availability is broad: Hydrogen comes to all Google AI Studio plans, the Gemini API, and the Gemini app.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.