Gemini 4 Hydrogen is generally available
TL;DR
On November 12, 2026 Google made Gemini 4 Hydrogen generally available as gemini-4-hydrogen. The API has a free tier and paid pricing of $1.50 per million input tokens and $7.50 per million output tokens. It adds an xhigh thinking level, defaults to high, and is the default model for the Antigravity agent in Gemini Managed Agents.
- Status
- Live
- Released
- Plans
- Free and paid Gemini API; Pro and Ultra for Antigravity and Sheets
What changed
Google's API release notes for November 12, 2026 mark gemini-4-hydrogen generally available. Google describes it as bringing its current intelligence to everyone, with speed and efficiency for everyday coding and agentic workflows. It is on the free tier and at paid-tier pricing of $1.50 per million input tokens and $7.50 per million output tokens. It adds a fourth thinking level, xhigh, next to low, medium, and the default high. It is now the default model for the Antigravity agent in Gemini Managed Agents.
The same day's developer guide is the spec:
- Free and paid tiers in Google AI Studio and the Gemini API, on both GenerateContent and the Interactions API. The free tier is rate-limited. The guide points at the rate limits page for the numbers.
- Paid price is flat across the 1 million token context window: $1.50 per million input tokens and $7.50 per million output tokens, including thinking tokens. Implicit caching discounts cached input by 95%, to $0.075 per million tokens.
- Thinking levels are
low,medium,high(the default), andxhigh. The guide says to start athighand reservexhighfor hard multi-step problems.xhighuses more tokens and adds latency, and it can overthink a simple prompt. - Input, thinking, and output tokens share one 1,048,576-token window. A synchronous response, streaming or not, stops at 65,536 tokens. Longer output sets
backgroundto true, or uses client-side continuation. Ifbackgroundis true andmax_output_tokensis unset, the server stops after 65,536 tokens. - Google recommends it as the upgrade from Gemini 3.8 Flash and from earlier Gemini 3 and 2.5 models, for everyday coding, interactive web apps, tool-using agents, multimodal documents, and parallel subagent work. Carbon stays the frontier model. Argon stays the lower-token choice for most advanced long-horizon work.
The November 12 announcement from Koray Kavukcuoglu says Hydrogen also rolls out beyond the API. Developers build with it in the Gemini API in AI Studio and Android Studio. Paid Google AI Pro and Ultra subscribers can try it in Google Antigravity. Enterprises deploy it in Gemini for business, in Google Antigravity, and in Gemini in Google Sheets for eligible business subscriptions. In the Gemini app it is in the model dropdown, and free users get it through auto model selection when it is the model for the prompt. It is in Google Search for all users. Gemini in Google Sheets starts with Google AI Pro and Ultra subscribers.
The same post introduces Orchestrator in Google Antigravity, which switches between models such as Carbon and Hydrogen in one session. On Terminal-Bench 4.0, Google says Orchestrator reached near-Carbon task success while using 24% less quota and finishing 19% faster, on average. For parallel agents, Carbon can coordinate while Hydrogen runs worker and verifier agents.
On the move from Gemini 3.8 Flash, the guide says Hydrogen finishes tasks 2.5 times faster at the default high setting, more than triples Terminal-Bench 4.1 scores (57% versus 18.2%), and cuts fabricated claims by more than 5x by flagging missing context instead of guessing. The announcement also says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at the default high thinking level, against 50 for Argon, and 54 at xhigh.
Why it matters
The model string is gemini-4-hydrogen. A request copied from Gemini 3.8 Flash, or from Carbon or Argon, needs a few edits before it behaves the way the guide describes.
From gemini-3.8-flash and earlier Gemini 3 models, the guide's checklist is:
- Change the model string to
gemini-4-hydrogen. - Treat the context window as one shared 1,048,576-token budget for input, thinking, and output. Gemini 3.8 Flash keeps a separate 1 million input budget and a 64K output budget. Hydrogen's single-request cap is still 65,536 tokens. Remove a hardcoded low
max_output_tokens, such as 8,192, or raise it to 65,536. To spend fewer tokens, lowerthinking_level. - Set
thinking_level. Hydrogen defaults tohigh. It rejectsminimalwith an error. Carbon and Argon default tomediumand cap athigh. - Remove explicit caching.
cachedContentsandcaches.createare not supported, andcached_contentfields come off the request. Implicit caching runs by default on prompts of at least 4,096 tokens. - Remove
temperature,top_p, andtop_k. The server ignores custom values if they are still sent. - Declare tools with function calling, or constrain JSON with structured outputs. Asking for raw-text JSON tool calls in the system prompt, with no tools declared, can return
MALFORMED_FUNCTION_CALL. - Keep conversation state. The Interactions API uses
previous_interaction_id. A client-managedgenerateContenthistory sends prior model turns and thought signatures back unchanged. A thought signature is not copied across parallelfunctionCallparts. - For output past 64K tokens, set
backgroundto true and setmax_output_tokensabove 65,536, or use client-side continuation in the latest SDKs. - Billing counts every token the model processes, including thinking tokens and, when you use grounding, internal grounding system instructions and function declarations. Grounding with Google Search stays billed per query.
Switching from Carbon or Argon needs no request-schema change. Set the model string, and set thinking_level when you want medium or xhigh. Hydrogen also allows up to 65,536 tokens per non-streaming hop on REST and SDK calls. Carbon and Argon allow 32,768 tokens per non-streaming REST hop. The guide says you can send a planning turn to gemini-4-carbon and an execution turn to gemini-4-hydrogen in one session with previous_interaction_id.
The Antigravity agent inside Gemini Managed Agents now defaults to this model. That is separate from the Antigravity app, where the announcement says paid Google AI Pro and Ultra subscribers can try Hydrogen.
Frequently asked questions
What is the model ID and the price?
gemini-4-hydrogen. The free tier is free of charge and rate-limited. On the paid tier, Google's guide says $1.50 per million input tokens and $7.50 per million output tokens, including thinking tokens, across the full 1 million token window. Cached input from implicit caching is $0.075 per million tokens.
How do I control how much it thinks?
Set thinking_level to low, medium, high, or xhigh. high is the default. xhigh is the new level, for harder multi-step work, and it uses more tokens.
What breaks if we swap it in for Gemini 3.8 Flash?
Change the model string, drop minimal and explicit caches, and drop temperature, top_p, and top_k. The default thinking level is high, not the medium default on Carbon and Argon. Output past 65,536 tokens needs background execution or client-side continuation. The guide's migration checklist is the full list.