API Many새 모델: Qwen 3.8 Max, Claude Haiku 5.5, Claude Opus 5.5, Claude Sonnet 5.5
GeminiLLMGuide

Gemini 4 Hydrogen: what it is, how thinking levels work, and how to call it

Gemini 4 Hydrogen is Google’s fast Gemini 4 model and the recommended upgrade from Gemini 3.8 Flash. Thinking levels up to xhigh, one shared 1M-token window, pricing, and an OpenAI-compatible example.

2026. 11. 13. · API Many

Google released Gemini 4 Hydrogen on November 12, 2026, as the fast member of the Gemini 4 family. Where Gemini 4 Carbon and Argon are built for long, deep reasoning, Hydrogen is built for quick execution in everyday coding and agent workflows, and Google names it the recommended upgrade from Gemini 3.8 Flash and earlier 3.x and 2.5 models. It is generally available on the Gemini API in both the free and paid tiers.

Sources:Google: Introducing Gemini 4 Hydrogen (Nov 12, 2026) · Gemini API: What's new in Gemini 4 Hydrogen

Gemini 4 Hydrogen at a glance

Gemini 4 HydrogenSource
ReleasedNovember 12, 2026 (generally available)Google
Model IDgemini-4-hydrogenModel page
Input / outputText, image, video, audio and PDF in; text outModel page
Context window1,048,576 tokens shared by input, thinking and outputModel page
Default output limit65,536 tokens per response; more with Long decodeLong decode
Thinking levelslow, medium, high (default), xhighThinking guide
Google’s standard price$1.50 / $7.50 per 1M input / output tokens; thinking billed as outputGemini API pricing
Not supportedImage or audio generation, Live API, explicit context cachingModel page

Thinking levels, including xhigh

Hydrogen thinks at high by default. You can lower it to medium or low for faster, cheaper replies, or raise it to xhigh, a level only Hydrogen offers among Gemini 4 models. Google reserves xhigh for multi-step coding and agent tasks where high is not enough; on simple extraction it can overthink and spend the output budget. minimal is not supported. Through the OpenAI-compatible API the level is the reasoning_effort field.

Sources:Gemini API: Thinking · Gemini API: OpenAI compatibility

  • low: the fastest replies, for routing, short answers and simple tool calls.
  • medium: more reasoning for everyday tasks.
  • high (default): Google’s recommended balance of quality and speed.
  • xhigh: the deepest reasoning, for hard multi-step coding and agent work. Leave room in max_tokens, because thinking tokens count toward it.

One 1M window for input, thinking and output

Unlike Gemini 3.8 Flash, which counts a 1M input window and a separate output limit, Hydrogen shares 1,048,576 tokens across input, thinking and output. If the input plus max_tokens would exceed the window, the output limit is cut to the space that is left. A response returns up to 65,536 tokens by default; Long decode goes further. Old code that caps output at 8,192 tokens is likely to cut answers off once thinking tokens are counted.

Sources:Gemini API: Gemini 4 Hydrogen model page · Gemini API: Long decode

What Google reports

  • Artificial Analysis Intelligence Index: 51 at the default high level, above Gemini 4 Argon’s 50, and 54 at xhigh.
  • Compared with Gemini 3.8 Flash at the default level: tasks finish about 2.5 times faster, Terminal-Bench 4.1 rises from 18.2% to 57%, and unsupported claims drop more than fivefold.

Source:Google: Introducing Gemini 4 Hydrogen (Nov 12, 2026)

These are Google’s statements in its launch post; API Many has not run its own benchmarks.

Call Gemini 4 Hydrogen with the OpenAI SDK

On API Many, Gemini 4 Hydrogen is served at the OpenAI-compatible /v1/chat/completions endpoint with the model ID gemini-4-hydrogen. Keep your OpenAI client, change the base URL and key, and pass reasoning_effort when you want a level other than high.

from openai import OpenAI

client = OpenAI(base_url="https://apimany.com/v1", api_key="YOUR_API_KEY")

reply = client.chat.completions.create(
    model="gemini-4-hydrogen",
    reasoning_effort="xhigh",  # low | medium | high (default) | xhigh
    messages=[{"role": "user", "content": "Find the bug in this function and propose a patch: ..."}],
)
print(reply.choices[0].message.content)
print(reply.usage)  # thinking tokens are included in completion_tokens

What you pay on API Many

Gemini 4 Hydrogen costs 입력 $1.875 · 출력 $9.375 on API Many. Thinking tokens are billed as output tokens, cached input is billed at the lower cached rate, and failed requests are not charged. Each response returns the token usage and the USD cost, and the same balance pays for every other model.

Hydrogen, Argon or Carbon?

  • Hydrogen: fast execution for everyday coding, agents, extraction and chat, and the upgrade path from Gemini 3.8 Flash.
  • Argon and Carbon: long-horizon, deep reasoning. Google suggests pairing them, for example Carbon to plan and Hydrogen as the worker or verifier.

Source:Gemini API: What's new in Gemini 4 Hydrogen

What is the Gemini 4 Hydrogen model ID?

gemini-4-hydrogen. On API Many, send it to POST /v1/chat/completions with your API Many key.

How do I use xhigh thinking?

Set reasoning_effort to "xhigh" in an OpenAI-compatible request. The other levels are low, medium and high (the default); minimal is not supported.

How large is the Gemini 4 Hydrogen context window?

1,048,576 tokens, shared by input, thinking and output. Each response returns up to 65,536 output tokens by default.

Are thinking tokens billed?

Yes. Thinking tokens are billed as output tokens, both by Google and on API Many.

Can Gemini 4 Hydrogen generate images or audio?

No. It accepts text, images, video, audio and PDFs but only returns text, and it does not support the Live API.

  1. Google: Introducing Gemini 4 Hydrogen (Nov 12, 2026) — Release date, positioning and Google-reported results
  2. Gemini API: Gemini 4 Hydrogen model page — Model ID, modalities, context window and unsupported features
  3. Gemini API: What's new in Gemini 4 Hydrogen — Migration notes from Gemini 3.8 Flash and how Hydrogen pairs with Carbon and Argon
  4. Gemini API: Thinking — Thinking levels, including xhigh
  5. Gemini API: Long decode — Default 65,536-token output limit and Long decode
  6. Gemini API pricing: Gemini 4 Hydrogen — Google’s standard, Batch and Priority prices
  7. Gemini API: OpenAI compatibility — reasoning_effort mapping in the OpenAI-compatible API
하나의 API 키로 체험하기

이 글의 모든 모델을 API Many에서 쓸 수 있습니다. $10부터 종량제.

모델 둘러보기 →

블로그 더 보기