When should I choose Carbon over Hydrogen?

Google positions Carbon for deep work on complex tasks and Hydrogen for speed and efficiency in everyday coding and agent workflows. Carbon defaults to medium thinking; Hydrogen defaults to high and adds xhigh. Compare completion quality, latency and total tokens on your workload rather than treating the name or token price as the cost of a task.

Can one request return a million output tokens?

The million-token figure depends on continuation across segments. A regular Carbon request is limited to 65,536 output tokens when streaming or 32,768 otherwise. Thinking, function calls and the answer share that allowance. RouteMux compatibility endpoints do not promise automatic Google long decode.

What changes when migrating from Gemini 3.x?

Review thinking levels, ignored sampling parameters and explicit cache usage. Messages does not accept an exact thinking.budget_tokens budget. URL images, document blocks and hosted tools are outside the current compatibility surface. Use Base64 images and preserve complete tool-call and tool-result history.

Gemini 4 Carbon

Save 50%

google/gemini-4-carbon

For complex coding, reasoning and multimodal tasks.

Get an API key View code sample

Live model playground

Try Gemini 4 Carbon here

Tune the request, see the current price ceiling, and inspect the real response without leaving this model page.

streamchat

1Configure input

Prompt

0 / 20,000

Add images

Maximum estimated cost—

2Live result

Your result will appear here

Enter an input on the left. The same RouteMux Playground execution and billing path powers this embedded test.

Context window: 1M
Max output: 1M
Input modalities: TextImage
Output modalities: Text
Capabilities: Vision · Reasoning · Tool calling · JSON mode · Coding
Upstream release: 2026-11-10
RouteMux name: google/gemini-4-carbon

What is Gemini 4 Carbon

Gemini 4 Carbon, released by Google on November 10, 2026, targets deep work on complex tasks. Google names coding, finance, legal work, cybersecurity and science among its intended domains. Use the model ID gemini-4-carbon when you want to evaluate sustained analysis, difficult reasoning or a workflow that repeatedly calls tools. Hydrogen is the sibling to compare when everyday coding, speed and efficiency matter more to the task.

Carbon has a single 1,048,576-token budget shared by input, thinking, function calls and the final answer. Its thinking levels are low, medium and high, with medium as the default. Custom temperature, top_p and top_k values are ignored. Moving from Gemini 3.x therefore requires checking sampling-dependent behavior and using thinking levels to vary reasoning effort.

RouteMux exposes OpenAI Chat Completions and Anthropic Messages with streaming, ordinary function tools and Base64 image input. Image understanding does not generate images. Office documents, audio, video and hosted tools available through Google's native APIs are separate from this compatibility surface. One Carbon request produces at most 65,536 output tokens when streaming, or 32,768 without streaming. Thinking consumes that output allowance. Google's million-token output figure requires continuation across segments, rather than one ordinary API response.

RouteMux input costs $1 per million tokens and output costs $5 per million tokens.

Before you choose

When should I choose Carbon over Hydrogen?: Google positions Carbon for deep work on complex tasks and Hydrogen for speed and efficiency in everyday coding and agent workflows. Carbon defaults to medium thinking; Hydrogen defaults to high and adds xhigh. Compare completion quality, latency and total tokens on your workload rather than treating the name or token price as the cost of a task.
Can one request return a million output tokens?: The million-token figure depends on continuation across segments. A regular Carbon request is limited to 65,536 output tokens when streaming or 32,768 otherwise. Thinking, function calls and the answer share that allowance. RouteMux compatibility endpoints do not promise automatic Google long decode.
What changes when migrating from Gemini 3.x?: Review thinking levels, ignored sampling parameters and explicit cache usage. Messages does not accept an exact thinking.budget_tokens budget. URL images, document blocks and hosted tools are outside the current compatibility surface. Use Base64 images and preserve complete tool-call and tool-result history.

Gemini 4 Carbon API pricing

Prices verified 2026-11-14

InputOfficial $2 / M tokens

$1/ M tokensSave 50%

OutputOfficial $10 / M tokens

$5/ M tokensSave 50%

Cache read $0.05/M · Cache write —/M

Estimate cost

Input tokensOutput tokens

Estimated cost$0.015

Key capabilities

Reasoning for complex work: Google positions Carbon for deep work across several domains. Compare low, medium and high on your own tasks instead of applying one thinking setting everywhere.
One shared token window: Long source files, tool history, thinking and answers consume the same 1,048,576-token window. More input leaves less room for reasoning and output.
Tools and image understanding: Chat Completions and Messages accept ordinary function tools and Base64 images, allowing screenshot observations and tool results to participate in multi-turn workflows.
Implicit caching, fixed sampling: Google offers implicit caching for eligible prompts of at least 4,096 tokens. Custom temperature, top_p and top_k values are ignored.

Common use cases

Complex software engineering: A core use named by Google: reason over substantial code, plan changes through multiple tool steps and check the results.
Professional analysis: Google includes finance and legal work among its intended domains. Analyze supplied material with domain experts reviewing the conclusions.
Science and security research: Google also names science and cybersecurity, including analysis that needs several reasoning steps and supporting tools.

Limited time10% extra Credits on your first top-up — up to 200 bonus Credits

Applies to your first top-up within 72 hours of signing up. Minimum $20, once per account.

Interface and protocols

OpenAI Chat CompletionsStreamingAnthropic MessagesStreaming

Use any of these IDs to call this model via the API.

RouteMux namegoogle/gemini-4-carbon

Replace the ROUTEMUX_KEY placeholder with your API key. Create one →

from openai import OpenAI

client = OpenAI(
    base_url="https://api.routemux.com/v1",
    api_key="<ROUTEMUX_KEY>",
)

completion = client.chat.completions.create(
    model="google/gemini-4-carbon",
    messages=[{"role": "user", "content": "What is the meaning of life?"}],
)
print(completion.choices[0].message.content)

base_url: https://api.routemux.com/v1

Performance (last 30 days)

4 requests · last 30 days

Total usage1.9K

P50 first-token latencyInsufficient data (4 requests)

Avg throughputInsufficient data (4 requests)

Success rate78.3%

Uptime · 24htested via openai_chat

78.26%

Success rate · 60d · from real requests

Gemini 4 Carbon FAQ

›What is Gemini 4 Carbon?

For complex coding, reasoning and multimodal tasks.

›How large is the context window of Gemini 4 Carbon?

Gemini 4 Carbon supports a context window of up to 1M tokens, with up to 1M output tokens per request.

›How much does the Gemini 4 Carbon API cost?

Through RouteMux, Gemini 4 Carbon costs $1 for input and $5 for output (USD per 1M tokens), below the official list price.

›How do I call Gemini 4 Carbon via API?

Gemini 4 Carbon is OpenAI-compatible: point your base URL at RouteMux and set the model field to google/gemini-4-carbon — no code changes needed.

›Which input and output modalities does Gemini 4 Carbon support?

Gemini 4 Carbon accepts text, image as input and produces text as output.

›When was Gemini 4 Carbon released?

Gemini 4 Carbon was released on November 10, 2026.

Sources

View as Markdown