Google launches Gemini 4 Hydrogen, 2.5 times faster and open to free users
Google launched the latest model of the Gemini 4 series, Hydrogen, which emphasizes that response speed is increased by 2.5 times, errors are reduced by five times, and it supports multi-model collaboration and millions of token contexts.
Quick Look
- Google DeepMind launched the Gemini 4 Hydrogen model on November 12, focusing on faster speed and lower cost.
- The new model's task speed is 2.5 times faster than the previous generation, errors are reduced by five times, and it supports millions of token contexts and multi-model collaboration.
- It has been gradually rolled out to Gemini Apps and developers around the world.
AI-generated summary
Why It Matters
Google launched Hydrogen, its third model after Gemini 4 Argon and Carbon, focusing on daily program development and data analysis.
Google DeepMind launched the latest model of the Gemini 4 series, Gemini 4 Hydrogen, on November 12, focusing on faster response times and lower usage costs.
Google claims that under default settings, the new model can complete tasks about 2.5 times faster than the previous generation Gemini 3.8 Flash, and the occurrence of error messages has been reduced by more than five times.
Hydrogen has begun to be imported into Gemini App, Google search and developer platforms, and free users also have the opportunity to use it.
The Hydrogen is the third model launched after the Gemini 4 Argon and Carbon.
Compared with the first two models, which focus on complex reasoning and long-term research work, Hydrogen is more suitable for daily program development, organizing files, analyzing data, and operating AI tools.
The Terminal-Bench 4.1 program development test released by Google shows that Hydrogen scored 57%, which is significantly higher than Gemini 3.8 Flash's 18.2%, but still lower than Claude Sonnet 5.5's 63.9%.
The third-party evaluation agency Artificial Analysis also announced the test results. Hydrogen scored 51 points in the default high thinking mode, slightly higher than the 50 points of Gemini 4 Argon; if it is switched to the highest xhigh mode, the score can be increased to 54 points.
However, deeper thinking also requires more time. In the test, it took about 6 minutes to complete the task in high mode on average, and it increased to about 13 minutes in xhigh mode.
Users can choose between four modes: low, medium, high and xhigh according to the difficulty of the work.
Hydrogen supports text, image, audio and video input, and has a context capacity of up to 1 million tokens, which can process a large number of documents or codes at one time.
API pricing is $1.50 per million input tokens and $7.50 per million output tokens, maintaining the same unit rate even when using longer contexts.
Google has also added a multi-model collaboration function to the AI development tool Antigravity, letting Carbon be responsible for more complex planning, while Hydrogen takes over program writing and testing.
Officials claim that this division of labor can reduce credit consumption by about 24% and shorten task completion time by 19%.
At present, Hydrogen has begun to be launched to Gemini App and Google search users around the world. The free version can be automatically selected by the system, while paying users can obtain different model choices and quotas according to the subscription plan.
Developers can use it through the Gemini API, Google AI Studio, and Google AI Pro and Ultra subscribers can also choose it in Antigravity.
Google has not announced the discontinuation of Gemini 3.8 Flash, and existing users can still choose a suitable model according to their actual needs.
Open Questions
- When will Google officially announce the discontinuation of Gemini 3.8 Flash?
- Will the usage limit be adjusted for free and paid users in the future?




