Key takeaways
- Google set Gemini 4 Hydrogen API pricing at $1.50 per million input tokens and $7.50 per million output tokens, flat across its 1 million token context window.
- Google said Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default setting and finishes tasks 2.5 times faster than Gemini 3.8 Flash.
- The model began rolling out globally on Nov. 12 in the Gemini app, Google Search, AI Studio, the Gemini API and Gemini for business.
In this story 5 sections
Google's new Gemini 4 Hydrogen model costs $1.50 per million input tokens and $7.50 per million output tokens through the Gemini API, at one flat rate across its full 1 million token context window, Alphabet (NASDAQ: GOOGL) said on Nov. 12. The company presents Hydrogen as the quick, lower cost member of the Gemini 4 family, aimed at routine coding and agent jobs, while the larger Gemini 4 Carbon and Gemini 4 Argon take on long and difficult reasoning work.
Hydrogen is also the model that puts Gemini 4 in front of Google's widest audience. Google said the global rollout started on Nov. 12 across the Gemini app, Google Search, Google AI Studio, the Gemini API and Gemini for business, plus the Google Antigravity coding tool for paying Google AI Pro and Ultra subscribers.
The figures Google published
These results come from the announcement post by Koray Kavukcuoglu, senior vice president of Google DeepMind. USA Publishing Group has not tested them independently.
- Artificial Analysis Intelligence Index: 51 at the default high thinking setting, one point above Argon's 50, and 54 at the new xhigh setting.
- Compared with Gemini 3.8 Flash: jobs complete 2.5 times faster at the default level, and the Terminal-Bench 4.1 score climbs to 57% from 18.2%. Google added that fabricated claims dropped by more than five times.
- Partner testing: early partners measured Hydrogen running 2 to 2.8 times faster than 3.8 Flash on data-heavy enterprise tasks while spending 12% to 23% fewer tokens.
- Other benchmarks: 58.8% on HLE-Diamond, 52.3% on the Vals Tax Agent Bench, where Google says it ranks first, and 55.8% on the Vals Legal Research Bench, a tie for first.
- Context size: up to 1 million tokens in and up to 1 million tokens out.
Our own math on the price list: each output token costs five times as much as an input token. A request that sends 200,000 tokens and receives 20,000 back would come to $0.30 for input plus $0.15 for output, or $0.45 in total.
Four thinking levels and an automatic model switch
Developers can run Hydrogen at low, medium, high or xhigh thinking levels, with high as the default. Google says the steps rise in a predictable way, so a team can match quality, wait time and cost to the depth of reasoning a given job actually needs.
Google also added a feature called Orchestrator to Antigravity. Instead of keeping one model for an entire session, it hands work back and forth between Carbon and Hydrogen. On Terminal-Bench 4.0, Google said Orchestrator came close to Carbon's success rate while using 24% less quota on average and finishing 19% sooner. For setups with several agents running in parallel, Carbon can act as the coordinator while Hydrogen handles the agents that write and check code.
What changes in the Gemini app and Search
In the Gemini app, Hydrogen shows up in the model menu. Free users reach it through automatic model selection when the app decides it fits a prompt, according to a note at the bottom of Google's post. Google says the model sticks more closely to requested tone and format and does a better job calling tools such as Gmail, Google Drive and connected apps like Airtable and Dropbox.
Google said Hydrogen now drives Search responses for all users worldwide. The company named three changes: better handling of long, multi-part or misspelled questions, more visual answers, and an upgraded query fan-out method that points people to a broader set of links. Google did not say how many additional sites that will surface. The Search update lands days after Google's November core update.
Safety claims
Measured against 3.8 Flash, Google said Hydrogen cuts overreach on agent tasks by as much as 81% and shows 12 times fewer misaligned actions and shortcuts in tool use and coding tests. The company reported a 96.7% score on its internal agent honesty evaluation and said the model resists indirect prompt injection more than 2.3 times better on the Gray Swan IPI benchmark. Google pointed readers to the Hydrogen model card for full details.
Where Hydrogen is available
- Developers: the Gemini API in AI Studio and Android Studio, and Antigravity for AI Pro and Ultra subscribers.
- Businesses: Gemini for business through the API or the model menu, Antigravity, and Gemini in Google Sheets on eligible plans.
- Consumers: the Gemini app model menu and Google Search, with Gemini in Sheets starting with AI Pro and Ultra subscribers.
For the pricing of the bigger model, see our earlier report on Gemini 4 Carbon pricing and availability. More coverage is on our AI page.
Share this story
Spotted an error? Tell us


