Gemini 4 Hydrogen is here and everyone can use it

Abstract:

Just now, Google launched Gemini 4 Hydrogen (hydrogen), focusing on speed and execution efficiency

.

Hydrogen is directly open to everyone

Google search is also connected to Hydrogen, which can better understand complex, scattered and even misspelled questions, and provide more intuitive answers and more relevant web links.

According to Google, in the default High inference setting, Hydrogen can complete tasks 2.5 times faster than the previous generation Gemini 3.8 Flash. Its command following ability has been further improved, and the number of fabricated facts has been reduced by more than 80%.

On the Artificial Analysis Intelligence Index list, it even surpasses the higher-end Gemini 4 Argon.

Google search first used the new model

Robby Stein, vice president of Google search products, called this access the largest model upgrade of Google search to date. Gemini 4 Hydrogen has begun to be rolled out to users around the world, covering all languages ​​supported by AI mode.

The most direct change is that you can spend less time thinking about search keywords in the future.

Now, you can type the ideas in your head directly into the search box. Even if the question is long, has many conditions, or even contains spelling errors, Hydrogen can more accurately understand what you are looking for and organize the answer according to these requirements.

Search results have also become more flexible.

Google said that the upgraded search can choose a more appropriate presentation method based on the question.

When information needs to be compared, it can generate tables; when encountering complex knowledge questions, it can help users understand the answers through pictures, visual content and even interactive tools.

Some things that previously required opening multiple websites and repeatedly comparing information can now be solved directly on the search page.

At the same time, Google Search has also upgraded Query fan-out (query expansion) technology. In the face of complex problems, the system can retrieve multiple relevant directions at the same time, cover a wider range of information sources, and provide more relevant web links.

This is actually quite interesting. AI search has been criticized for leaving answers directly on the search page, making it less and less necessary for users to click on external websites. Now Google is beginning to emphasize that the upgraded AI search can help users discover more websites.

Google itself is also facing pressure to monetize AI search. In an earnings conference call on October 28, Google Chief Commercial Officer Philipp Schindler admitted that the monetization efficiency of AI Mode is lower than that of traditional search, and its query volume has grown rapidly, which has also dragged down the growth of search advertising revenue.

As for whether the website can regain more clicks after the number of links increases, it will take time to prove.

Gemini is more obedient and more capable

Maybe Google feels that the codename Hydrogen is intuitive enough. Although there is no Flash in the name, Hydrogen assumes the role of the Gemini 4 series of high-efficiency models.

Google positions Carbon and Argon for complex deep reasoning tasks, while Hydrogen puts more emphasis on execution efficiency, especially programming, tool calling, and multi-step tasks.

But this model, which focuses on speed and execution efficiency, even exceeds its own high-end models.

On the Intelligence Index list of the third-party evaluation agency Artificial Analysis, Hydrogen scored 51 points in the High inference gear, and further increased to 54 points in the Xhigh gear, surpassing the 50 points of Gemini 4 Argon.

According to the official Benchmark released by Google, Hydrogen performs particularly well in knowledge work, automation and reasoning.

In Agents' Last Exam v1.1, which examines AI's ability to handle complex knowledge tasks, it achieved a score of 71.6%, surpassing Claude Sonnet 5.5's 60.4% and GPT-6.1 Terra's 69.8%.

In the Vending-Bench 2 test, which simulates long-term business tasks, Hydrogen finally accumulated a fund balance of $14,549, also ahead of the other two models.

In terms of programming, it led with 13.5% in SWE-sweep, but fell behind Claude Sonnet 5.5's 63.9% in Terminal-Bench 4.1 with 57%.

Besides running scores, the more direct change is that Gemini is more obedient and capable in daily use.

Google said that Hydrogen can better adhere to tone, format and specific rules, maintain task objectives across multiple rounds of conversations, and reduce the number of repeated revisions.

When processing complex data, it can extract information from uploaded files, organize tables and visual content, and complete tasks in conjunction with Gmail, Google Drive, and connected applications such as Airtable and Dropbox.

Hydrogen's tool calling capabilities have also been improved. Google said that the model can more accurately select the tools that need to be used, and can adjust subsequent operations even if the tool call fails or the search returns no results.

In an official demonstration, Hydrogen used Google Antigravity to generate multi-page slides directly in Google Slides based on a YouTube lecture video and design template. Subsequently, it also independently checked the presentation and fixed typography issues such as overlapping text.

There is another change worth noting: This time, Google launched a new Orchestrator feature for Google Antigravity.

Antigravity has previously supported multiple agents working together using different models. The newly launched Orchestrator can automatically switch between Carbon and Hydrogen according to the execution progress of the same task.

According to the workflow given by Google, Carbon can first analyze the task, formulate a plan, and then hand it over to Hydrogen to modify the code, update dependencies and run tests. If new difficulties are encountered during execution, the task can be returned to Carbon for analysis and adjustment.

The entire process does not require users to manually switch models.

Google said that in internal tests, Orchestrator achieved a task success rate close to that of using Carbon alone, while reducing credit consumption by an average of 24% and increasing task completion speed by 19%.

This feature is currently available to Google AI Ultra users, and will support more models and subscriptions in the future.

In terms of price, Hydrogen’s API charges US$1.5 per million input tokens and US$7.50 per million output tokens, which is lower than Gemini 4 Argon’s current preferential prices of US$2 per million input and US$10 per million output.

Compared with the previous generation Flash, the price of the new model has increased. The current input and output prices of Gemini 3.8 Flash are US$0.75 and US$3.75 respectively, which is only half of Hydrogen's.

However, the preferential price for the old version of Flash will end on December 31, and will be adjusted to the same level as Hydrogen from January 1 next year.

Netizen: What about Argon?

After Hydrogen was released, people started testing it immediately.

Netizen @SPAC89 showed off his Three.js project made with Hydrogen Extra. He said that the entire project took less than two hours, of which nearly 80% was spent on rendering, consuming only 5% of the weekly quota of the Google AI Pro subscription.

He is quite satisfied with the new model, and even said that this is the first time he feels that the $20 Pro subscription has almost no limit.

User @StatsWire conducted a test with four Gemini models: Let them use Three.js to create a 3D Mario game that can actually be played without calling any external 3D models or textures.

Some netizens in the comment area expressed the hope that the author could also announce the time and cost required for each model to generate the game. The author stated that these data will be supplemented in the next test.

Google search for the news of access to Hydrogen has also attracted a lot of discussion.

Some netizens thought of a very interesting scenario: "Developers in 2026: AI wrote a bug. Now you throw the error message into Google search, and it is still the same AI that explains the bug."

However, some netizens are not optimistic about Google's search business.

These complaints are not directed at Hydrogen, but more about dissatisfaction with Google's existing products.

Under the official Google DeepMind post, Hydrogen’s performance also raises another question: So what is the use of Argon now?

After all, Hydrogen is cheaper and faster, and in some reviews it surpassed Argon, which has a higher positioning.

Some people also question the speed increase - what is the use of light speed increase?

One netizen pointed out that a 2.5x speed increase is indeed good, but if the answer is wrong, the cost of re-executing the task may be higher than a slow but correct answer.

Another netizen joked,

AI can now write code so fast that it can finish it before finishing the coffee itself, but checking the code may require drinking the entire remaining pot.

The number one on the model ranking list is always eye-catching, and everyone is happy to discuss who has surpassed whom. But when it comes to using AI every day, people often have another set of evaluation criteria.

Netizen @mattyapro’s comment reflects the sentiments of many people:

“For me, the models I use every day are more important than the models that top the leaderboards.”

It's great that the day-to-day work is getting attention.