What is Gemini 4 Hydrogen?
Gemini 4 Hydrogen is the fast, cheaper tier of Google DeepMind's Gemini 4 family, released on 12 November 2026. The model card says it is adapted from Gemini 4 Carbon. Google aims it at everyday coding, agent work and multimodal tasks, and leaves long-horizon reasoning to Carbon and Argon.
The model ID is gemini-4-hydrogen, listed as stable. It reads text, images, video, audio and PDF files and writes text. Input and output share a window of 1,048,576 tokens; one synchronous response is capped at 65,536 tokens by default, and long decode stretches output to the full window. Thinking has four levels: low, medium, high (the default) and a new xhigh. The knowledge cutoff is March 2026.
API pricing is $1.50 per million input tokens and $7.50 per million output tokens, thinking included, flat across the whole window. Cached input costs $0.075, Batch and Flex are half price and Priority costs 80% more. Unlike Carbon and Argon, it has a free tier in AI Studio, where prompts are used to improve Google's products.
How much does Gemini 4 Hydrogen cost?
| Plan | Price | What you get |
|---|---|---|
| Standard | $1.50 in / $7.50 out per 1M tokens | Cached input $0.075; same rate across the 1M window |
| Batch and Flex | $0.75 in / $3.75 out per 1M tokens | Paid tier only |
| Free tier | Free | AI Studio and Gemini API; no Batch, Flex or grounding; prompts used to improve Google products |
API: $1.50 input / $7.50 output per 1M tokens, flat to the full 1M window; cached input $0.075; Batch and Flex $0.75 / $3.75; Priority $2.70 / $13.50. Free tier in AI Studio (data used to improve Google products). Antigravity: paid Google AI Pro or Ultra.
Compare this price with other models
Is Gemini 4 Hydrogen worth it?
Hydrogen is the cheapest way into Gemini 4 and, on the evidence so far, gives up little for it. Artificial Analysis scores it 51 on its Intelligence Index at the default high thinking level, one point above Gemini 4 Argon, and 54 at xhigh, 16th of 81 models. It costs $1.50 per million input tokens and $7.50 per million output, against $4 and $20 for Carbon and Argon from 1 December.
It writes quickly but starts slowly. Artificial Analysis measures about 170 output tokens a second at every thinking level, but 16.35 seconds to the first answer token at high and 44.27 seconds at xhigh, against 6.25 seconds at low. It is also verbose: at xhigh it spent 304 million output tokens on the index, 79th of 80 models, or $3.16 per index task. Google says that at high it finishes tasks 2.5 times faster than Gemini 3.8 Flash, and early partners report runs 2 to 2.8 times faster with 12% to 23% fewer tokens.
Google's model card compares it with Claude Sonnet 5.5 and GPT-6.1 Terra and gives it 7 of 10 rows, including 58.8% on HLE-Diamond, 71.6% on Agents' Last Exam v1.1 and 76.1% on AutomationBench-AA. Sonnet 5.5 leads on the rows closest to day-to-day agent work: Terminal-Bench 4.1 (63.9% against 57.0%), OSWorld-2.1 computer use (83.9% against 77.6%) and Terminal-Bench-Science. These are Google's runs, and some tests, such as Multimodal Knowledge Agent (53.4%), are Google's own. Against Gemini 3.8 Flash, Terminal-Bench 4.1 rises from 18.2% to 57%.
In Google Antigravity, Hydrogen is for paid Google AI Pro and Ultra subscribers and powers Orchestrator, which passes one task back and forth between a larger model that plans and repairs and Hydrogen, which edits files and runs tests. Google says Orchestrator came close to Carbon's success rate on Terminal-Bench 4.1 with 24% less quota and 19% less time. Google's launch post also shows Hydrogen in Antigravity turning Google Sheets into an animated pixel canvas by shrinking thousands of cells, painting scenes with Python, =LET formulas and conditional formatting.
The catch: nearly every benchmark is Google's, and independent agent results will take weeks. The model card reports 5.57 points more unjustified refusals than 3.8 Flash, mostly on multimodal prompts. Only implicit caching is supported, and there is no Live API. Until 31 December, 3.8 Flash costs half as much per token; from January the two cost the same.
Our take: for routine coding and agent steps on the Gemini API, Hydrogen is the sensible default, and Carbon becomes the exception a task escalates to when it stalls.
Where it earns its place
A developer on a paid Google AI Pro plan using Antigravity
Make the monthly quota last through a week of feature work
Choose Orchestrator in the model menu (the Argon version on Pro, Carbon on Ultra): the larger model plans and Hydrogen does the edits and test runs. Google measured 24% less quota and 19% faster runs than Carbon alone on Terminal-Bench 4.1, at close to Carbon's success rate.
A platform team running Gemini 3.8 Flash in production agents
Decide whether to switch before Flash's introductory price ends
From January Hydrogen costs the same as 3.8 Flash, $1.50 / $7.50, and Google reports Terminal-Bench 4.1 at 57% against 18.2% and more than five times fewer fabricated claims. Change the model ID to gemini-4-hydrogen and check prompts that relied on explicit caching, which it does not support.
A tax research team at a mid-sized accounting firm
Run research agents over long case files without a long-context surcharge
Google reports 52.3% on the Vals Tax Agent Bench and 55.8% on the Vals Legal Research Bench. A 500K-token file with a 4K-token answer costs about $0.78 at standard rates and $0.39 on Batch, because the price does not rise past 200K tokens.
Before you commit
- Setup
- Model ID gemini-4-hydrogen in AI Studio, the Gemini API or Android Studio, on every tier including free. Thinking level is low, medium, high (default) or xhigh. In Antigravity it appears in the model menu for paid Google AI Pro and Ultra subscribers as the rollout reaches them.
- Integrations
- Function calling, Google Search and Maps grounding (paid tier; 5,000 free requests a month shared across Gemini 3 and later, then $14 per 1,000), computer use and Batch. No Live API. Also in Gemini for business, the Gemini app (free users only through automatic model selection), Google Search, Sheets for AI Pro and Ultra, and Antigravity's Orchestrator and SDK.
- Data and privacy
- Free-tier prompts and responses are used to improve Google's products; paid-tier data is not, except in the invite-only Contributor mode. The model card says Hydrogen reaches the alert thresholds for CBRN and cyber uplift but not the critical levels, and that misuse safeguards were strengthened.
- Cost at scale
- 10,000 document reviews a month at 200K tokens in and 4K out cost about $3,300 at standard rates and $1,650 on Batch or Flex. Thinking bills as output, and Artificial Analysis counts $3.16 per index task at xhigh, so the thinking level moves the bill more than the list price.
- Limits
- 1,048,576 tokens shared by input, thinking and output; 65,536 output tokens per synchronous response by default, more with long decode. Search grounding is capped at 1,500 requests a day. The model card lists occasional slowness or timeouts and higher token use at high thinking levels.
What’s good
- 51 on Artificial Analysis at default high, above Argon; 54 at xhigh
- $1.50 / $7.50 per 1M tokens, flat across the 1M window; Batch and Flex half price
- Free API tier, unlike Carbon and Argon
- About 170 output tokens a second; text, image, video, audio and PDF input
What’s not
- Trails Claude Sonnet 5.5 on Terminal-Bench 4.1 and OSWorld-2.1, even on Google's table
- Very verbose at xhigh: 304M output tokens on the index, 79th of 80
- 16-second wait for the first answer token at high, 44 seconds at xhigh
- Implicit caching only, no Live API; more unjustified refusals than 3.8 Flash
Best for
Gemini API and Antigravity users who want Argon-class results for routine coding and agent steps at well under Carbon's price.
Gemini 4 Hydrogen alternatives
Gemini 4 Hydrogen: common questions
- Is Gemini 4 Hydrogen worth it?
- Hydrogen edges Argon on Artificial Analysis for under 40% of Argon's December price, so it should be the default Gemini model for routine work; Claude Sonnet 5.5 still wins on terminal and computer-use tasks. We rate it 4.4 out of 5.
- How much does Gemini 4 Hydrogen cost?
- API: $1.50 input / $7.50 output per 1M tokens, flat to the full 1M window; cached input $0.075; Batch and Flex $0.75 / $3.75; Priority $2.70 / $13.50. Free tier in AI Studio (data used to improve Google products). Antigravity: paid Google AI Pro or Ultra.
- What are the downsides of Gemini 4 Hydrogen?
- Trails Claude Sonnet 5.5 on Terminal-Bench 4.1 and OSWorld-2.1, even on Google's table; very verbose at xhigh: 304M output tokens on the index, 79th of 80; 16-second wait for the first answer token at high, 44 seconds at xhigh; implicit caching only, no Live API; more unjustified refusals than 3.8 Flash.
- Who is Gemini 4 Hydrogen best for?
- Gemini API and Antigravity users who want Argon-class results for routine coding and agent steps at well under Carbon's price.
- How much does Gemini 4 Hydrogen cost for a team or in production?
- 10,000 document reviews a month at 200K tokens in and 4K out cost about $3,300 at standard rates and $1,650 on Batch or Flex. Thinking bills as output, and Artificial Analysis counts $3.16 per index task at xhigh, so the thinking level moves the bill more than the list price.
- Is Gemini 4 Hydrogen safe for company data?
- Free-tier prompts and responses are used to improve Google's products; paid-tier data is not, except in the invite-only Contributor mode. The model card says Hydrogen reaches the alert thresholds for CBRN and cyber uplift but not the critical levels, and that misuse safeguards were strengthened.
- What are the best Gemini 4 Hydrogen alternatives?
- The closest alternatives we have reviewed are Claude Sonnet 5.5, GPT-6.1 Terra, Gemini 3.8 Flash and Gemini 4 Carbon.
Answers checked Nov 16, 2026.