AI News

Google Releases Gemini 4 Hydrogen Model, Available to Free Users Too

Phoenix TechnologyUpdated Nov 13 08:19

AI summary

Google has released Hydrogen, a new model in the Gemini 4 series, featuring faster responses, lower cost, and stronger AI Agent execution capabilities, and it is integrated into platforms including the Gemini App, Google Search, Google AI Studio, and Antigravity; free users of the Gemini App can also use it through model selection. Artificial Analysis evaluations show that Hydrogen scores 51 on the intelligence index in the default High tier, surpassing Gemini 4 Argon's 50, and rises to 54 when the newly added xhigh tier is enabled; API pricing is $1.5 per million input tokens and $7.5 per million output tokens, with support for a 1 million token context window. Google has also launched the Orchestrator feature for Antigravity, which can automatically switch between models such as Carbon and Hydrogen within the same session based on the task; the company says it maintains success rates close to Carbon on Terminal-Bench 4.0 while reducing quota consumption by an average of 24% and increasing task completion speed by 19%.

Why read this

SelectedAI score 85

The original text provides Hydrogen's pricing, reasoning tiers, and multi-platform availability, which can be used to assess the position of low-cost models in Agent workflows.

Topics

Tags

Body · AI translation

Just now, the new Gemini 4 model has arrived again, and at last it's available to everyone. ifanr Download the app for an exclusive first look. November 13, 2026, 07:11:11, from Beijing

Before we've even recovered from the shock of Jensen Huang resigning, Google has released a brand-new model: Gemini 4 Hydrogen.

According to the official positioning, Carbon and Argon are mainly aimed at complex reasoning, long-term planning, and highly difficult software engineering.

Hydrogen, by contrast, focuses on faster response speeds, lower usage costs, and more efficient AI Agent execution capabilities, and is better at quickly writing code, calling tools, and handling everyday tasks that require continuous execution.

What's more interesting is that although Hydrogen is positioned as a high-efficiency model, some of its test results have already surpassed Gemini 4 Argon, and even come close to flagship products such as Claude Sonnet 5.5 and GPT-6.1 Sol.

At the same time, unlike the previous performances with no actual product, Google announced that Hydrogen will be integrated into multiple platforms including the Gemini App, Google Search, Google AI Studio, and Antigravity, and free users can also begin to experience the latest capabilities of the Gemini 4 series.

Performance surpasses Argon, speed increased by up to 2.5 times

Although its positioning is different, Hydrogen's actual performance is quite formidable.

According to evaluations by Artificial Analysis, Hydrogen scores 51 points at the default High reasoning level, surpassing Gemini 4 Argon's 50 points; after enabling the new xhigh level, the intelligence index further rises to 54 points.

This result is only 1 point lower than the scores of Claude Sonnet 5.5 and GPT-6.1 Sol at their highest reasoning levels.

What is more noteworthy is the price.

Hydrogen's API pricing is $1.5 per million input tokens and $7.5 per million output tokens, with a uniform price maintained across the full 1 million token context window.

Image |

According to Artificial Analysis's calculations, the per-task cost of Hydrogen at the xhigh level is about 32% of that of Claude Sonnet 5.5 at its highest level, though it still reaches about 2.5 times that of GPT-6.1 Sol.

The new model offers four reasoning intensity levels: Low, Medium, High, and xhigh, allowing users to choose among intelligence level, speed, and cost according to actual needs.

Among them, High is the default level, while xhigh is the newly added highest reasoning level.

In third-party evaluations, the per-task cost of Hydrogen at the xhigh level is about $3.16, while at the High level it drops to $1.87, with an intelligence index difference of only 3 points.

For most everyday tasks, the default level may already be sufficient, and there is no need to use the highest reasoning intensity every time.

In addition to cost, Hydrogen has also made clear progress in speed.

Data released by Google shows that, compared with Gemini 3.8 Flash, Hydrogen completes tasks at the default High level up to 2.5 times faster.

In the Terminal-Bench 4.1 test, the score rose from 18.2% to 57%, and fabricated information was reduced by more than 80%. Artificial Analysis also measured that Hydrogen's output speed is about 200 tokens per second, and at some reasoning levels it is already on the Pareto frontier of intelligence level and task latency.

Of course, Hydrogen does not lead in every item.

In the Terminal-Bench 4.0 test, it scored 54%, lower than Claude Sonnet 5.5's 64% and GPT-6.1 Sol's 56%.

By comparison, Hydrogen performs more prominently in visual reasoning and reducing hallucinations. Artificial Analysis's AA-Omniscience test shows that its hallucination rate is only 8%, while Claude Sonnet 5.5 and GPT-6.1 Sol are 43% and 52%, respectively.

In addition, Hydrogen supports text, image, audio, and video input, has a 1 million token context window, and supports up to 1 million tokens of output, and can handle materials such as complex PDFs, spreadsheets, charts, and long videos.

With a single prompt, Hydrogen can build a whole city

Benchmark scores are one thing, and the actual experience is another. After the model was released, developers quickly began to showcase Hydrogen's generation capabilities, especially in 3D scenes and interactive web pages.

Netizen @ai_for_success used Hydrogen with Swarm in Antigravity to create a 3D harbor called Porto Astrolabio. The demo includes a complete navigation interface and can switch between day and night, letting users freely explore the virtual port.

Another netizen, @LuminaBench, used the highest xhigh reasoning tier to create a 3D seaplane demo and commented that Hydrogen's performance is now comparable to Argon, while being faster and cheaper.

Netizen @kimmonismus's work is a Three.js futuristic city that can be freely explored in flight.

According to the author, he used the High tier and, after one follow-up prompt, generated procedural skyscrapers, rain effects, and traffic shuttling above the city.

Netizen @abhinavflac also showcased a 3D miniature building-block scene made with Hydrogen, which likewise demonstrated the model's ability to generate 3D content.

After seeing these netizens' works, I also prepared several sets of prompts to try using Hydrogen to tackle different types of interactive content.

I had Hydrogen create a little ghost adventure game on a late-autumn farm, requiring the character to explore an abandoned cabin, investigate objects, and interact with the surrounding environment.

Then I switched to an animated-style countryside cycling game, riding a bicycle full of fresh flowers through rice paddies, with the ability to freely switch between first-person and third-person views.

I even did a stunt with Jensen Huang, having Hydrogen use HTML and SVG to create a 2D animation of Jensen Huang riding a bicycle.

Finally, I tried a whitewater rafting game called Riverbound, in which four characters ride a red inflatable boat through valley rapids, testing water-flow physics and collision interactions.

Google also prepared several interesting examples.

For example, in a project called Geminisland, users only need to enter a single natural-language prompt to generate buildings in real time on a shared voxel island and explore it in first-person view.

Another demo showed Hydrogen working with Antigravity's front-end design Skill to generate a complete e-commerce website in one go, including a WebGL2 terrain shader, scroll-driven storytelling, 3D product cards, and a real-time product configurator.

Beyond web development, Hydrogen can also turn thousands of cells in Google Sheets into dynamic pixel art, or automatically create presentations in Google Slides based on YouTube talk videos and design templates.

Even more interestingly, after generating slides, the model also automatically checks for issues such as text overlap and layout anomalies, and tries to fix them.

From these cases, it is clear that Hydrogen's strengths are gradually extending from single-shot generation to continuous execution, tool calling, and autonomous checking—and these are exactly the capabilities AI Agents need in real work.

English text is being prepared.

Along with the release of Hydrogen, Google Antigravity also gained a new feature called Orchestrator.

But different tasks have different requirements for models.

Complex system design requires a flagship model like Carbon that excels at deep reasoning, while large amounts of code modification, testing, and verification work can be handed to Hydrogen, which responds faster and costs less.

Orchestrator can automatically switch between different models such as Carbon and Hydrogen within the same session according to the specific task, letting each take on the work it is best suited for.

For example, Carbon can be responsible for formulating the overall development plan, while multiple Hydrogen Agents are respectively responsible for writing code, running tests, and checking results, and can execute tasks in parallel in isolated workspaces.

According to Google's published Terminal-Bench 4.0 test results, Orchestrator maintains a success rate close to that of Carbon executing tasks alone, while reducing quota consumption by an average of 24% and improving task completion speed by 19%.

For developers, in the future they will only need to state the task requirements, and the system will automatically select models according to different work stages, reducing cost and waiting time while ensuring completion quality.

And this automatic division of labor further explains why Google launched an efficient model like Hydrogen.

Free users can use it too, and Google Search gets a major upgrade

In addition to developer tools, another important change in this Hydrogen release is that the Gemini 4 series is beginning to enter consumer products on a large scale.

According to Google's published rollout plan, Hydrogen is being released to users worldwide, covering the Gemini App, Google Search, Google AI Studio, the Gemini API, and enterprise services.

Among them, Hydrogen is available to users at all tiers in Google AI Studio, and free users of the Gemini App can also gain access through the model selection feature. Hydrogen in Google Antigravity is available to paid subscribers of Google AI Pro and Ultra.

For ordinary users, Hydrogen focuses on improving complex instruction understanding, multi-turn dialogue, and tool-calling capabilities.

For example, users can ask Gemini to find emails in Gmail and then combine them with materials in Google Drive to generate a report. The model can select the appropriate tools according to the task, reducing repeated explanations and manual operations.

Google has also strengthened the model's ability to follow writing tone, format, and specific rules, making content such as emails and research reports closer to users' actual requirements.

At the same time, Google Search has also received what Google calls its largest model upgrade to date.

Robby Stein, Vice President of Search products, said that Hydrogen can more accurately understand complex search intents, and even if a question contains multiple conditions, spelling errors, or is expressed in a relatively casual way, it can identify the information the user truly needs.

In addition, Google has upgraded its Query Fan-out technology, allowing the search system to simultaneously launch multiple related queries around a complex question and provide a broader range of reference websites. Search results are no longer limited to web links; they can also generate images, comparison tables, structured information, and even customized interactive tools based on the question.

For ordinary users who habitually use Google Search, these changes may be more intuitive than the scores on model leaderboards.

In just six weeks, Google has launched three Gemini 4 models in succession.

The emergence of Hydrogen also shows that when a cheaper, faster model has already surpassed Argon in some evaluations, how much work remains irreplaceable for an expensive flagship model?

Moreover, Google also has a huge product ecosystem such as Google Search, Gmail, and Google Drive, which can bring Hydrogen directly into ordinary users' daily work.

Perhaps Hydrogen is the most easily underestimated model in the Gemini 4 series. Flagship models keep proving what AI can do, while Hydrogen has a better chance of determining how many people can actually use these capabilities and at how low a cost.

"Special statement: The above content (including videos, pictures, and audio if any) is uploaded and published by users of Dafeng Hao, a self-media platform under Phoenix Net. This platform only provides information storage space services.

Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."

Source

Phoenix Technology