Gemini 4 Hydrogen: Pricing, Benchmarks, vs. Argon

Gemini 4 Hydrogen (gemini-4-hydrogen) is Google’s fast, efficient Gemini 4 model for everyday coding and agentic workflows, released on November 12, 2026. It is generally available with a free tier and standard paid API pricing of $1.50 per 1 million input tokens and $7.50 per 1 million output tokens across a unified 1,048,576-token context window.

What Is Gemini 4 Hydrogen?

Gemini 4 Hydrogen is a multimodal large language model from Google DeepMind designed as the speed- and cost-efficiency tier of the Gemini 4 family. Koray Kavukcuoglu, Senior Vice President at Google DeepMind and Chief AI Architect at Google, introduced the model in an official Google blog post alongside an announcement from Google CEO Sundar Pichai on X. While Gemini 4 Carbon and Gemini 4 Argon serve as Google’s frontier models for deep, long-horizon reasoning, Google positions Gemini 4 Hydrogen as the default workhorse for interactive coding, tool-using agents, and multimodal document tasks.

Google’s documentation names Gemini 4 Hydrogen as the recommended upgrade path from Gemini 3.8 Flash as well as earlier Gemini 3 and Gemini 2.5 models. In the Gemini API changelog for November 12, 2026, Google confirmed that gemini-4-hydrogen is generally available on the free tier and at paid-tier pricing of $1.50 per 1M input tokens and $7.50 per 1M output tokens, introduces a fourth xhigh thinking level, and is now the default model powering the Antigravity agent in Gemini Managed Agents. According to Google’s What’s new in Gemini 4 Hydrogen documentation, the model is available in both the GenerateContent API and the Interactions API.

Technical Specifications and Supported Capabilities

According to the official Gemini 4 Hydrogen model page and DeepMind model card as of November 13, 2026, the model carries the following core specifications:

Stable model ID: gemini-4-hydrogen (latest update: November 2026).
Input modalities: Text, image, video, audio, and PDF.
Output modality: Text only.
Unified context window: 1,048,576 tokens shared across input, thinking, and output tokens.
Maximum output tokens: 65,536 tokens per synchronous request by default, expandable up to the full 1,048,576-token window using long decode via background execution in 64K-token hops or client-side continuation.
Supported API features: Function calling, structured outputs, code execution, computer use, file search, Google Search and Google Maps grounding, URL context, configurable thinking (low, medium, high, and a new xhigh tier), implicit context caching, Batch API, Flex processing, and Priority processing.
Unsupported features: Live API, explicit context caching (cachedContents / caches.create), image generation, and audio generation.

Where You Can Access Gemini 4 Hydrogen

Google rolled out Gemini 4 Hydrogen across developer, enterprise, and consumer surfaces on launch day:

Developers: Available through the Gemini API in Google AI Studio and Android Studio. Paid Google AI Pro and Google AI Ultra subscribers can also use it inside Google Antigravity.
Enterprises: Accessible in Gemini for business via the API or the model drop-down menu, within Google Antigravity, and in Gemini in Google Sheets for eligible business plans.
Consumers: Available to all users in Google Search and via the model selector in the Gemini app. Free-tier Gemini app users are also routed to Hydrogen through automatic model selection when their prompt fits the model’s profile. Gemini in Google Sheets access is rolling out starting with Google AI Pro and Ultra subscribers.

How Much Does Gemini 4 Hydrogen Cost?

Gemini 4 Hydrogen costs $1.50 per 1 million input tokens and $7.50 per 1 million output tokens on the standard paid API tier as of November 13, 2026, alongside a rate-limited free tier. Unlike tiered pricing structures that charge more for long prompts, Google charges a flat rate across the entire 1,048,576-token context window.

The table below lists every pricing tier from the official Gemini API pricing documentation in USD per 1 million tokens:

Tier	Input (per 1M tokens)	Output incl. thinking (per 1M tokens)	Cached input (per 1M tokens)	Data used to improve Google products?
Free tier	Free	Free	Free (rate-limited)	Yes
Standard (paid)	$1.50	$7.50	$0.075	No
Batch	$0.75	$3.75	$0.0375	No
Flex	$0.75	$3.75	$0.0375	No
Priority	$2.70	$13.50	$0.135	No

Billing Rules for Thinking, Caching, and Search Grounding

When estimating your monthly spend or comparing options in our AI API pricing comparison, keep four billing mechanics from Google’s pricing documentation in mind:

Every token is billed: Output token charges include both final response tokens and internal thinking tokens. When you enable grounding tools, internal tool system instructions and function declarations are also billed as tokens.
Automatic 95% implicit caching discount: Explicit caching APIs are disabled on Gemini 4 Hydrogen, but implicit prompt caching runs by default on prompts of at least 4,096 tokens. Cache hits reduce the input price by 95%, down to $0.075 per 1 million tokens on the standard tier and $0.0375 per 1 million tokens on Batch and Flex.
Google Search grounding fees: Paid accounts receive a shared pool of 5,000 free Google Search grounding requests per month across Gemini 3 and newer models. After exhausting that pool, Search grounding costs $14 per 1,000 requests with a daily cap of 1,500 requests. Search grounding is not available on the free tier.
Free tier privacy: Prompts and outputs on the free tier are used to improve Google products, whereas paid tier traffic is not.

How Does Gemini 4 Hydrogen Compare to Argon, Carbon, and Gemini 3.8 Flash on Price?

Gemini 4 Hydrogen sits at $1.50 input and $7.50 output per 1 million tokens, matching the standard list price that Gemini 3.8 Flash will reach on January 1, 2027, while undercutting Gemini 4 Argon and Carbon by roughly 2.7x once their promotional pricing expires on December 1, 2026.

Because Google currently has two separate scheduled price changes in effect as of November 13, 2026, looking only at today’s sticker prices can be misleading when planning Q4 2026 and Q1 2027 budgets. Here is how the four models compare on the standard paid tier:

Model	Standard Input (per 1M tokens)	Standard Output (per 1M tokens)	Free Tier Available?	Key Pricing Dates
Gemini 4 Hydrogen	$1.50	$7.50	Yes	Flat rate in effect from Nov 12, 2026
Gemini 4 Argon	$2.00 (through Nov 30, 2026) $4.00 (from Dec 1, 2026)	$10.00 (through Nov 30, 2026) $20.00 (from Dec 1, 2026)	No	Prices double on Dec 1, 2026
Gemini 4 Carbon	$2.00 (through Nov 30, 2026) $4.00 (from Dec 1, 2026)	$10.00 (through Nov 30, 2026) $20.00 (from Dec 1, 2026)	No	Prices double on Dec 1, 2026
Gemini 3.8 Flash	$0.75 (through Dec 31, 2026) $1.50 (from Jan 1, 2027)	$3.75 (through Dec 31, 2026) $7.50 (from Jan 1, 2027)	Yes	Prices double on Jan 1, 2027

Worked Monthly API Cost Example

To see what these dates mean in practice, consider a workload that processes 10 million input tokens and 2 million output tokens per month on the standard paid tier (excluding cache hits and differences in thinking-token volume, since thinking tokens are billed as output):

Gemini 4 Hydrogen: (10 × $1.50) + (2 × $7.50) = $15.00 + $15.00 = $30.00 per month.
Gemini 4 Argon or Carbon (from December 1, 2026): (10 × $4.00) + (2 × $20.00) = $40.00 + $40.00 = $80.00 per month (about 2.7x the per-token cost of Hydrogen).
Gemini 3.8 Flash (through December 31, 2026): (10 × $0.75) + (2 × $3.75) = $7.50 + $7.50 = $15.00 per month (rising to $30.00 per month on January 1, 2027, matching Hydrogen).

If you are evaluating ways to reduce AI API costs, notice that Gemini 3.8 Flash is half the per-token price of Hydrogen only until the end of 2026. Furthermore, Google reports that early enterprise partners saw Hydrogen use 12% to 23% fewer tokens than 3.8 Flash on data-heavy tasks, which could narrow the effective gap for similar workloads; test on your own prompts.

How Does Gemini 4 Hydrogen Perform on Google’s Benchmarks?

According to Google’s launch blog post, Gemini 4 Hydrogen outperforms Gemini 3.8 Flash on speed, agentic coding, and factual accuracy, while edging out Gemini 4 Argon on the Artificial Analysis Intelligence Index. All performance figures below are vendor claims reported by Google on November 12, 2026.

Intelligence and Coding Benchmarks vs. Argon and 3.8 Flash

On the Artificial Analysis Intelligence Index, Google reports that Gemini 4 Hydrogen scores 51 at its default high thinking setting, compared to 50 for Gemini 4 Argon. Raising Hydrogen’s thinking parameter to the new xhigh tier pushes its Artificial Analysis score to 54.

When compared directly against Gemini 3.8 Flash, Google attributes several large gains to Hydrogen:

Terminal-Bench 4.1: Hydrogen scores 57%, up from 18.2% on Gemini 3.8 Flash.
Task completion speed: At its default high thinking setting, Google states that Hydrogen finishes tasks 2.5x faster than Gemini 3.8 Flash.
Enterprise partner testing: Google notes that early-access partners running graph, SQL, and data-heavy enterprise workflows observed Hydrogen running 2x to 2.8x faster while consuming 12% to 23% fewer tokens than 3.8 Flash.
Hallucinations and tool recovery: Google claims that Hydrogen reduces fabricated claims by more than 5x compared to 3.8 Flash and adapts more reliably when a tool call fails or a search query returns empty results.

Agentic, Domain, and Safety Evaluations

Google also published benchmark scores across specialized agent and safety suites:

Evaluation / Benchmark	Gemini 4 Hydrogen Result (Google claim)	Additional Context Reported by Google
Agent Arena (arena.ai)	Ranks #5 overall	arena.ai live benchmark for multi-step tasks
HLE-Diamond	58.8%	Leads its model class
Vals Tax Agent Bench	52.3%	Ranks #1 overall
Vals Legal Research Bench	55.8%	Ties for #1 overall
Multimodal Knowledge Agent (MKA)	53.4%	End-to-end agentic knowledge work over complex visual files
Agents’ Last Exam v1.1	71.6%	End-to-end professional workflows
Internal Agent Honesty Evaluation	96.7% honesty rate	Google internal safety and honesty test
Agentic Overreach Reduction	Up to 81% reduction	Google says overreach in agentic tasks drops by up to 81%
Gray Swan Indirect Prompt Injection	2.3x more resilient	Compared with Gemini 3.8 Flash

In addition to standalone benchmarks, Google introduced a new Orchestrator in Google Antigravity that dynamically switches between Gemini 4 Carbon and Gemini 4 Hydrogen within a single session. On Terminal-Bench 4.0, Google reports that this Carbon-plus-Hydrogen orchestrator achieved near-Carbon task success while consuming 24% less quota on average and finishing 19% faster.

How Do Gemini 4 Hydrogen’s Four Thinking Levels Work?

Gemini 4 Hydrogen introduces a fourth thinking level called xhigh alongside low, medium, and high, giving developers granular control over how many reasoning tokens the model spends before responding. While Gemini 4 Carbon and Argon default to medium and cap out at high, Gemini 4 Hydrogen defaults to high and supports xhigh.

According to Google’s What’s new in Gemini 4 Hydrogen documentation, each tier targets a distinct latency and reasoning profile:

low: Uses the fewest thinking tokens and delivers the lowest latency. Best suited for fast chat interfaces and high-throughput processing.
medium: Balances speed and reasoning for interactive coding, live UI prototyping, and latency-sensitive tool calls.
high (default): Google’s recommended baseline for the majority of production coding, analysis, and agentic tasks.
xhigh: Designed for hard, multi-step coding and agentic problems where high falls short. Google cautions that xhigh increases latency and token consumption and can overthink simple prompts.

Google advises engineering teams to run evaluation suites across all four thinking levels with a high max_output_tokens ceiling, then deploy the lowest thinking level that satisfies their quality bar.

What Changes When Migrating From Gemini 3.8 Flash to Gemini 4 Hydrogen?

Migrating from Gemini 3.8 Flash to Gemini 4 Hydrogen requires updating your model ID string and adjusting several request parameters because Hydrogen enforces stricter API defaults and uses a unified context window. If you are migrating from Gemini 4 Carbon or Argon instead, no request schema changes are required, and you can even route planning turns to Carbon and execution turns to Hydrogen within the same session (noting that Hydrogen supports 65,536 tokens per non-streaming REST hop versus 32,768 on Carbon and Argon).

When upgrading from gemini-3.8-flash, these are the main items from Google’s migration checklist:

Update the model identifier: Replace gemini-3.8-flash with gemini-4-hydrogen.
Raise low max_output_tokens caps: Gemini 3.8 Flash maintained a separate 1M input limit and 64K output limit, whereas Hydrogen uses a single unified 1,048,576-token window for input, thinking, and output combined. Remove hardcoded low max_output_tokens caps (such as 8,192) or raise them to 65,536 so thinking tokens do not truncate your visible output.
Update thinking level parameters: Hydrogen defaults to high thinking instead of 3.8 Flash’s medium default. If your code passes the legacy minimal thinking setting, remove it, as gemini-4-hydrogen rejects minimal with an API error.
Remove explicit context caching calls: Delete calls to cached_content or caches.create. Explicit caching is not supported; Hydrogen applies implicit caching automatically on prompts of 4,096 tokens or more.
Strip custom sampling parameters: Remove temperature, top_p, and top_k overrides from your requests. The server ignores custom sampling values on Gemini 4 Hydrogen.
Declare tools natively: Always register tools through native function calling or structured outputs. Prompting the model to emit raw-text JSON tool calls with zero declared tools can trigger a MALFORMED_FUNCTION_CALL error.
Preserve multi-turn state: Retain thought signatures or pass previous_interaction_id across multi-turn tool sessions so the model maintains its reasoning chain.

Which Gemini Model Should You Pick?

Choosing between Gemini 4 Hydrogen, Gemini 4 Argon, Gemini 4 Carbon, and Gemini 3.5 Flash-Lite comes down to how much deep reasoning your task requires versus your latency and budget constraints. If you are building a broader routing strategy, our guide on how to choose an AI model walks through the questions to ask before you commit.

Based on Google’s official model selection guidance as of November 2026, here is where each Gemini model fits:

Pick Gemini 4 Hydrogen for everyday software engineering, interactive web applications, tool-using agents, multimodal document workflows (text, PDF, audio, image, and video), and parallel subagents. It gives you Gemini 4 intelligence at $1.50/$7.50 per 1M tokens and supports 65,536 tokens per non-streaming hop.
Pick Gemini 4 Carbon for your hardest system architecture decisions, complex debugging, scientific research, and security audits. Carbon spends more thinking tokens per turn to tackle deep reasoning problems, and it pairs well with Hydrogen as the planning model in a two-model orchestrator setup.
Pick Gemini 4 Argon when you need frontier long-horizon task performance similar to Carbon’s class, but want lower token consumption per turn than Carbon. Keep in mind that both Argon and Carbon double in price from $2.00/$10.00 to $4.00/$20.00 per 1M tokens on December 1, 2026.
Pick Gemini 3.5 Flash-Lite when sub-second latency targets and the lowest possible per-token cost matter more than complex reasoning, such as high-volume classification and routing.

FAQ

Is Gemini 4 Hydrogen free to use?

Yes, Gemini 4 Hydrogen has a free tier in the Gemini API (via Google AI Studio) with rate limits, and consumer users can access it for free in Google Search and the Gemini app. On the API free tier, Google uses prompts and responses to improve its products, and Google Search grounding is disabled. Paid API tiers ($1.50 input / $7.50 output per 1M tokens) do not use your data to improve Google products.

Why is Gemini 4 Hydrogen more expensive than Gemini 3.8 Flash right now?

Through December 31, 2026, Gemini 3.8 Flash is priced at $0.75 per 1M input tokens and $3.75 per 1M output tokens, which is half of Gemini 4 Hydrogen’s $1.50/$7.50 standard price. On January 1, 2027, Gemini 3.8 Flash rises to $1.50 input and $7.50 output per 1M tokens per Google’s pricing page, making Hydrogen and 3.8 Flash identical in per-token price from that date forward.

Does Gemini 4 Hydrogen support context caching?

Gemini 4 Hydrogen supports automatic implicit context caching, which activates by default on prompts of at least 4,096 tokens and gives a 95% discount on cached input tokens ($0.075 per 1M tokens on the standard tier). However, Gemini 4 Hydrogen does not support explicit context caching (cachedContents or caches.create). If you are migrating from Gemini 3.8 Flash, you must remove cached_content fields from your API requests.

What is the maximum context window and output limit on Gemini 4 Hydrogen?

Gemini 4 Hydrogen features a unified 1,048,576-token context window that is shared across input tokens, internal thinking tokens, and output tokens. By default, synchronous API requests cap output at 65,536 tokens. For workflows that require longer generations, developers can enable long decode via background execution in 64K-token hops or client-side continuation to generate up to the full 1,048,576-token window.

Can Gemini 4 Hydrogen generate images or audio, or use the Live API?

No, Gemini 4 Hydrogen outputs text only. While it accepts text, images, video, audio, and PDF files as inputs, it does not support image generation, audio generation, or the bidirectional Live API. Its feature set is focused on text generation, code execution, computer use, structured outputs, function calling, file search, and grounding with Google Search and Google Maps.

Gemini 4 Hydrogen: Pricing, Benchmarks, and How It Compares to Argon and 3.8 Flash

What Is Gemini 4 Hydrogen?