Gemini 4 API: Carbon vs Argon vs Hydrogen
Compare Gemini 4 Carbon, Argon, and Hydrogen APIs on TTAPI. Explore coding and agent workflows, request formats, token pricing, and launch-rate deadlines.
What you will take away
- Match the model to the depth and pace of the work
- Compare TTAPI token prices and launch-rate deadlines
- Use the documented Gemini request formats
Carbon targets demanding, sustained reasoning. Argon brings frontier reasoning to complex everyday engineering and analysis. Hydrogen emphasizes fast execution for daily coding and agent workflows. Match the model to the work and compare total task cost, including output and thinking tokens.
Three models for different kinds of work
Gemini 4 Carbon, Argon, and Hydrogen are available on TTAPI under their exact Google model IDs. All three use TTAPI's Gemini channel with 40% off official token prices. They share an integration surface while serving different priorities in coding, reasoning, and agent execution.
Google positions Carbon for its hardest engineering and scientific problems, Argon for efficient frontier reasoning, and Hydrogen for fast everyday execution. A useful comparison starts with the tasks your application repeats, the depth of analysis they require, and how quickly a finished answer is needed.
Carbon, Argon, and Hydrogen at a glance
The same prompt can call for different kinds of effort. A difficult architectural decision benefits from sustained analysis, while a small implementation task needs a prompt answer and a clear way to verify it.
| Model | Primary focus | Example brief |
|---|---|---|
| Gemini 4 Carbon | Demanding reasoning and sustained multi-step work | Investigate a subtle cross-service failure and assess architectural fixes |
| Gemini 4 Argon | Efficient frontier reasoning for engineering and analysis | Plan a code migration with dependencies and acceptance checks |
| Gemini 4 Hydrogen | Fast everyday coding and agent execution | Implement a scoped change, run checks, and report the result |
Carbon and Argon for deeper engineering work
Carbon is Google's most capable Gemini 4 reasoning model. Its published focus includes complex software engineering, mathematical and scientific problem solving, and autonomous work across many steps. Consider it for a difficult investigation where the result depends on testing several hypotheses and reconciling evidence across a large body of work.
Argon is Google's efficient frontier reasoning option, designed for complex coding, analysis, and agent tasks with faster responses and lower token spend per turn than Carbon. Its developer introduction highlights code migrations and other workflows that require structural precision over many steps.
Carbon and Argon currently share the same standard token prices. Their cost for a completed task can still differ because the amount of generated and thinking content can differ. Compare the delivered result, elapsed time, and billed tokens on representative work.
Hydrogen for everyday coding and execution
Hydrogen emphasizes speed and efficiency in everyday coding and agent workflows. Google describes it as a model that can handle daily work on its own or carry out scoped tasks alongside Carbon and Argon. That makes it a candidate for repeated implementation, test execution, and verification steps in a larger workflow.
For a team building an API integration, an initial division of work could put a difficult design review on Carbon or Argon and a small, well-specified implementation on Hydrogen. Evaluate that division using your own acceptance criteria and usage records.
Token prices and the launch-rate deadlines
TTAPI applies its existing Gemini 40% discount to Google's official rates. The table shows standard-tier TTAPI prices in USD per million tokens during the published launch periods. Output billing includes thinking tokens, and cached input for these models uses implicit caching.
Carbon and Argon's launch rates run through November 30, 2026. From December 1, TTAPI's discounted input, cached-input, and output rates become $2.40, $0.12, and $12 per million tokens. Hydrogen's cached-input rate is $0.045 through December 31, 2026 and becomes $0.09 on January 1, 2027. Its input and output rates remain $0.90 and $4.50.
The same pricing notice also covers Gemini 3.8 Flash. Its TTAPI input, cached-input, and output rates are $0.45, $0.045, and $2.25 per million tokens through December 31, 2026, then $0.90, $0.09, and $4.50 from January 1, 2027. In each case, the channel discount continues when Google's promotional period ends.
| Model | Input / 1M | Cached input / 1M | Output / 1M |
|---|---|---|---|
| Gemini 4 Carbon | $1.2 | $0.06 | $6 |
| Gemini 4 Argon | $1.2 | $0.06 | $6 |
| Gemini 4 Hydrogen | $0.9 | $0.045 | $4.5 |
Carbon and Argon rates apply through November 30, 2026. Hydrogen's cache rate applies through December 31, 2026.
Use the documented Gemini request formats
The compatible Gemini endpoint accepts messages, model, and the optional stream flag. Use gemini-4-carbon, gemini-4-argon, or gemini-4-hydrogen as the model value. The request below shows the shape of a coding brief.
TTAPI also documents the native route at /v1beta/models/{model}:generateContent. That route puts the model ID in the URL and uses contents in the request body, with native fields for system instructions, tools, and model-supported generation settings. Keep the request structure paired with the route you choose.
curl --request POST \
--url 'https://api.ttapi.io/gemini/chat/completions' \
--header "TT-API-KEY: $TTAPI_KEY" \
--header 'Content-Type: application/json' \
--data '{
"model": "gemini-4-hydrogen",
"messages": [
{
"role": "user",
"content": "Review this migration plan and identify the three highest risks."
}
],
"stream": false
}'Native Generate Content requests and settings
The native endpoint places the model ID in the URL. Its required body field is contents, an array of conversation turns with parts. The example adds systemInstruction and generationConfig.maxOutputTokens using their documented native structure. Authentication still uses the TT-API-KEY header.
Optional native fields include tools, toolConfig, safetySettings, systemInstruction, generationConfig, cachedContent, serviceTier, and store. Generation settings include temperature, topP, topK, structured-output schemas, thinkingConfig, imageConfig, and other documented options. Their availability depends on the selected model. These options belong to the native request structure; the Chat Completions reference lists messages, model, and stream.
Each Gemini detail page provides separate request-format selectors for the parameter reference and code examples. Choose Generate Content in the section you are using to see the matching endpoint, path parameter, native body fields, and cURL, Node, or Python example.
curl --request POST \
--url 'https://api.ttapi.io/v1beta/models/gemini-4-hydrogen:generateContent' \
--header "TT-API-KEY: $TTAPI_KEY" \
--header 'Content-Type: application/json' \
--data '{
"contents": [
{
"role": "user",
"parts": [
{
"text": "Review this migration plan and identify the three highest risks."
}
]
}
],
"systemInstruction": {
"parts": [
{
"text": "Give concise, practical engineering guidance."
}
]
},
"generationConfig": {
"maxOutputTokens": 1024
}
}'