
Screenshots show the official pages on the capture date and may differ from the pages at publication. Select an image to enlarge it.
On November 12, 2026 (US time), Google released a new AI model, "Gemini 4 Hydrogen." Sitting alongside the top models Gemini 4 Carbon and Gemini 4 Argon, it is built for speed and low cost, and Google positions it for everyday programming and work done by AI agents. It succeeds Gemini 3.8 Flash, and Google recommends it as the model to switch to from 3.8 Flash and from earlier Gemini 3 and 2.5 models.
A big jump in performance over 3.8 Flash
According to Google, Hydrogen's main performance results are as follows.
- 57% on an evaluation of solving tasks through commands (Terminal-Bench 4.1), more than three times Gemini 3.8 Flash's 18.2%
- With the standard settings, it finishes work 2.5 times faster than 3.8 Flash, and the rate at which it makes up untrue content is less than one-fifth
- It scored 51 on the overall index from the third-party firm Artificial Analysis, above the higher-tier Argon (50). With its deepest thinking setting, it scored 54
- Google says it also ranks first in its class on agent evaluations of specialized work, such as tax and legal research
In the development tool Google Antigravity, Google says that switching between Carbon and Hydrogen depending on the task kept success rates close to Carbon's while cutting usage by 24% and shortening working time by 19%.
On safety, Google explains that compared with 3.8 Flash, it cut the rate at which agents take actions they were not asked to take by up to 81%, and the rate at which they go beyond what they were allowed to do by up to half.
API pricing and specs
The model ID in the Gemini API is "gemini-4-hydrogen," and it was released as a stable version (generally available).
- Pricing is $1.5 per 1M input tokens and $7.5 per 1M output tokens. The price is the same at any length up to about 1M tokens
- Repeated identical input is automatically discounted, and that portion costs $0.075 per 1M tokens
- It can also be used on the free tier, which has usage limits
- Input can be text, images, video, audio and PDFs. Output is text only
- Thinking depth now has four levels, "low," "medium" and "high," plus "xhigh," which thinks more deeply. The default is high
On the other hand, it does not support generating images or audio, or the Live API for real-time conversations. Google cautions that xhigh uses more time and tokens and can overthink simple questions.
Availability in the Gemini app and Search
Hydrogen is rolling out gradually worldwide from launch.
- Gemini app: added to the model choices. Free users can use it when the feature that automatically picks a model for each question selects Hydrogen
- Google Search: available to all users. Google calls it "the biggest model upgrade ever in Search," and says that across all languages AI Mode supports, it improves understanding of what complex questions are asking and answers that use tables and images
- For developers and businesses: Google AI Studio, Android Studio and Gemini for business. In Antigravity, paid Google AI Pro and Ultra subscribers can use it
Google has not said anything specific about the Japanese language or availability in Japan.
CoAI's take
Hydrogen is, in effect, the workhorse model behind the Gemini app and Search that many people use every day. Its API price is lower than the introductory price of the top-tier Carbon ($2 input, $10 output), and on some evaluations it beats Argon. Developers currently using 3.8 Flash should compare cost and speed on the same tasks before deciding whether to switch.
However, there are changes from 3.8 Flash, such as generation settings like temperature being ignored and values outside the allowed range causing errors. We recommend checking Google's migration guide before switching.