[go: up one dir, main page]

GPT Image 2.5 Flare & Sunburst are live on EvoLinkTry GPT Image 2.5

Gemini 4 Carbon API

Access Gemini 4 Carbon through EvoLink’s unified API for coding, multi-step reasoning and secure development evaluations. Reuse your API key with OpenAI-compatible Chat Completions or the Gemini native API, and choose a thinking level for each request.

GoogleText GenerationAvailable
From $1.900 / 1M input tokens$4.000 official priceIntro pricing-53%
API docs
Thinking levelsStreamingChat CompletionsGemini native API
Production routeLive
Context
Not published
Best For
Coding, reasoning and secure development
Input
Text
Output
Text

Choose Gemini 4 Carbon

Google's Gemini 4 Carbon, announced on October 28, 2026. Input, cached input, and output are metered separately; output includes thinking tokens, and current rates follow Google's introductory pricing.

Gemini 4 Carbon

Google's newest Gemini 4 model

Selected
Model ID
gemini-4-carbon
Best for

Evaluate code changes, multi-step reasoning and secure development tasks. Choose a thinking level, then compare accepted results, retries and total task cost with Gemini 4 Argon before switching production traffic.

Input
$1.900 / 1M-53%
129.2 cr / 1M$4.000official price
Cache read
$0.096 / 1M-52%
6.5 cr / 1M$0.200official price
Output
$9.500 / 1M-53%
646 cr / 1M$20.000official price

All rates are per 1M tokens, shown in USD and credits, and reflect your account's current pricing.

Gemini 4 Carbon pricing

Estimate what one Gemini 4 Carbon request costs before you integrate. The calculator uses your account's current rates, with official pricing as a reference.

Introductory pricing — the rates below follow Google's introductory price for Gemini 4 Carbon ($2.000 input / $10.000 output per million tokens). Google has not announced when it ends; after that, EvoLink's usual discount applies to Google's standard price ($4.000 input / $20.000 output per million tokens).

Request calculator

Enter the token mix for one request.

Estimated request cost

Gemini 4 Carbon
USD$0.0048
Credits0.323

Official estimate $0.0100 · save $0.0053 (52%)

Input tokens0.1292 cr
Cache read tokens0 cr
Output tokens0.1938 cr

Minimum charge: 0.01 credits per request.

Budget guide

Approximate requests using the current token mix.
Add credits
$10
About 2012 requests

For quick testing

$50
About 10526 requests

For regular development

$100
About 21052 requests

For production evaluation

What is Gemini 4 Carbon?

Google announced Gemini 4 Carbon on October 28, 2026, following Gemini 4 Argon, with a focus on sustained engineering tasks and cybersecurity defense. EvoLink provides text input and output, adjustable thinking levels and streaming through two API formats. Evaluate code changes and reasoning results against your own acceptance criteria before expanding a workload.

Gemini 4 Carbon

Gemini 4 Carbon specs and capabilities

Numbers come from the EvoLink route configuration; capabilities are what the API exposes today.

Input
Text
Output
Text
Reasoning
Configurable reasoning effort
Protocols
Chat Completions · Gemini native API
Model ID
gemini-4-carbon

Gemini 4 Carbon for coding, reasoning and secure development

Use relevant code and technical context, define the result you need, and compare quality, response time and total task cost.

Gemini 4 Carbon for coding and code review

Provide the relevant files, error details and expected behavior to investigate a bug, review a change or draft a multi-file fix. Run the tests and review the diff; compare accepted fixes, unintended changes and manual rework. Repository access and test execution belong to your application or Agent.

Gemini 4 Carbon for multi-step reasoning

Break a design decision or technical question into constraints, candidate solutions and checks. Choose low, medium or high thinking per request, then compare answer quality, elapsed time and billed output, including thinking tokens. Keep the same task and acceptance rules when comparing levels.

Gemini 4 Carbon for secure development

Use authorized code and issue reports to evaluate vulnerability analysis, patch candidates and regression-test suggestions. Check that a proposed fix addresses the issue while preserving valid behavior. Keep human review and external test results in the acceptance process; an explanation alone does not establish that code is secure.

Two ways to use Gemini 4 Carbon: EvoLink API or Agent

Use the EvoLink API for product backends and batch jobs, or call Gemini 4 Carbon from Codex, Claude, or Gemini for coding and analysis workflows. Both paths share the same EvoLink API key, balance, model ID, and request history.

Option 1

Integrate with the EvoLink API

Best for: product backends, batch jobs, automated pipelines

Send Gemini native generateContent requests to EvoLink and control the model ID, system instruction, output budget, tools, and structured output.

  1. 1Create an EvoLink API key in the console
  2. 2Call the generateContent endpoint on the EvoLink base URL with the model ID shown above in the request path
  3. 3Send one representative request and read usageMetadata for input, cached, and output tokens
  4. 4Set maxOutputTokens and retries per task; keep function-call parts and their results across turns
Option 2

Call it with an Agent

Best for: coding, review, and analysis tasks in Codex, Claude, and Gemini

Give the Agent the task, the inputs to include, and the acceptance criteria. It assembles the request, calls Gemini 4 Carbon through EvoLink, and returns the answer with token usage.

  1. 1Set EVOLINK_API_KEY in your local environment; never put it in code or a prompt
  2. 2Describe the task, the inputs to include, and the expected output format
  3. 3Ask the Agent to call Gemini 4 Carbon through EvoLink and show the request before sending
  4. 4Let the Agent report the answer, token usage, and any error body

Gemini 4 Carbon API code example and error handling

This example shows the shortest runnable request: a Gemini native generateContent call with a system instruction, a user message, and an output budget. Open the API tab for the complete parameter and response reference.

View complete API docs
cURL
# Gemini native API
curl -X POST "https://direct.evolink.ai/v1beta/models/gemini-4-carbon:generateContent" \
  -H "Authorization: Bearer $EVOLINK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "systemInstruction": { "parts": [{ "text": "You are a careful reviewer. Explain your reasoning briefly." }] },
    "contents": [
      { "role": "user", "parts": [{ "text": "Review the design below and list the three biggest risks:\n<design notes>" }] }
    ],
    "generationConfig": {
      "maxOutputTokens": 8192,
      "thinkingConfig": { "thinkingLevel": "medium" }
    }
  }'

# thinkingLevel accepts low, medium, or high. Gemini 4 Carbon thinks by
# default, so use low for tasks that do not need extended reasoning.
# For streaming, call streamGenerateContent?alt=sse with the same body.
# Output tokens, including thinking tokens, are billed at the output rate.
# The response includes candidates[0].content.parts and usageMetadata
# (prompt, thinking, and output token counts).

Invalid request or unsupported parameter

Check the model ID, the contents array, and parameter ranges against the API reference; remove fields this route does not support.

Authentication or balance issue

Check the Authorization bearer token and confirm the available balance in the console.

Context length exceeded

Prompt tokens exceed the Gemini 4 Carbon context window. Trim or retrieve only the relevant evidence and reuse cached prefixes.

Rate limited (429)

Back off and retry with jitter; batch or queue requests instead of sending parallel bursts.

Content or tool call rejected

Review sensitive content, malformed tool-call arguments, and JSON schema mismatches before retrying.

Gemini 4 Carbon API integration checks

Keep one EvoLink client and API key. These four steps cover what is specific to Gemini 4 Carbon.

01

Set the model ID

For Chat Completions, set model to "gemini-4-carbon" in the request body. For the native API, use that ID in /v1beta/models/{model}:generateContent. The two protocols have different request shapes.

Required
02

Choose a thinking level

Gemini 4 Carbon thinks by default. In Chat Completions, set reasoning_effort to low, medium, or high; on the native API, set thinkingLevel in thinkingConfig to low, medium, or high. Use low when a task does not need extended reasoning: thinking tokens count as output and are billed at the output rate.

Cost control
03

Budget the output

Google has not published the maximum length of a single Carbon response. The native API accepts maxOutputTokens up to 1,048,576 as a request budget, not a guaranteed response length. Output includes thinking tokens. Set maxOutputTokens (max_tokens in Chat Completions) for each task, and separately cap the whole Agent task, including retries.

Cost control
04

Compare before you switch

Record task ID, model, thinking level, acceptance result, retries, billed usage and elapsed time for Carbon and the model you run today. Use the same inputs and acceptance rules, and keep a tested rollback route such as Gemini 4 Argon.

Evaluate

Thinking is on by default

Gemini 4 Carbon thinks even on short prompts, so a simple request can still use dozens of thinking tokens. They are billed at the output rate ($9.500 per 1M tokens): Chat Completions counts them in completion tokens, and the native API reports them separately in usageMetadata. Use the low level for tasks that do not need extended reasoning, and check thinking tokens before you scale a workload.

Context window and output limit

Google has not published Gemini 4 Carbon’s context window or the maximum length of a single response. The native API accepts maxOutputTokens up to 1,048,576 as a request budget; this does not guarantee that many generated tokens. Output includes thinking tokens. Set a budget for each task and test representative long inputs before depending on them.

Gemini 4 Carbon vs Gemini 4 Argon and Gemini 3.8 Flash

EvoLink

Prices are the current EvoLink rates for each model. Use the table to shortlist, then compare results and total cost per task on your own workload.

Gemini 4 Carbon
Input / output$1.9 / $9.5
ContextNot published
CachingCached input rate
Best forCoding, reasoning and secure development
Gemini 4 Argon
Input / output$1.9 / $9.5
Context1.05M
CachingAutomatic cache reads
Best forThe previous Gemini 4 model, with long-output continuation on the native API.
Gemini 3.8 Flash
Input / output$0.714 / $3.564
Context1.05M
CachingAutomatic cache reads
Best forThe newest Flash generation for stronger coding, reasoning, and agent workloads.

Gemini Model Family

Same API key and balance — switch tiers without changing your integration.

Compare all Gemini models
Gemini 4 Hydrogen

Gemini 4 Hydrogen

Google’s lower-priced Gemini 4 model, with thinking levels, image input, function calling, and Google Search.

View model
Gemini 4 Argon

Gemini 4 Argon

Google’s Gemini 4 model, with a 1M-token output limit and a lower rate for cached input.

View model
Gemini 3.8 Flash

Gemini 3.8 Flash

The newest Flash generation for stronger coding, reasoning, and agent workloads.

View model

Other text models on EvoLink besides Gemini 4 Carbon

Compare these models on the same task, acceptance checks and tool setup. Check each model page for supported features and current pricing.

Claude Fable 5.2

Claude Fable 5.2

Compare code review and complex reasoning with Claude Fable 5.2 on the same acceptance checks.

View model
Claude Opus 5.5

Claude Opus 5.5

Evaluate Claude Opus 5.5 on engineering tasks and compare accepted results with total cost.

View model
GPT-6.1 Terra

GPT-6.1 Terra

Compare GPT-6.1 Terra on reasoning quality, response time and cost per completed task.

View model
Kimi K3.1 Pro

Kimi K3.1 Pro

Evaluate Kimi K3.1 Pro on repository and document tasks using the same input and acceptance rules.

View model

Gemini 4 Carbon guides and related reading

Gemini 4 Carbon announced: what changes after Argon

Gemini 4 Carbon announced: what changes after Argon

Explore Google’s announcement, Carbon vs Argon, and the methods and limitations behind the published benchmarks.

Read guide

Gemini 4 Carbon API FAQ

What is Gemini 4 Carbon best used for?

Evaluate Carbon for coding and code review, multi-step reasoning and secure development tasks. Supply the relevant text, define acceptance checks, and measure accepted results, manual rework and total task cost. Your application or Agent supplies repository access and test execution.

Is the Gemini 4 Carbon API available through EvoLink?

Yes. Gemini 4 Carbon is available on EvoLink. Use your existing EvoLink API key and send model "gemini-4-carbon".

What model ID should I use for the Gemini 4 Carbon API?

Use "gemini-4-carbon". For Chat Completions, set it as model in the request body; for the native API, use it in /v1beta/models/{model}:generateContent. The two protocols have different request shapes.

How much does the Gemini 4 Carbon API cost on EvoLink?

$1.900 per 1M input tokens, $9.500 per 1M output tokens, and $0.096 per 1M cached input tokens. Output includes thinking tokens. The discounts shown on this page are measured against Google's standard price of $4.000 input and $20.000 output per 1M tokens, not the introductory rate.

What is Google's introductory pricing for Gemini 4 Carbon, and when does it end?

Google launched Gemini 4 Carbon at the same introductory price as Gemini 4 Argon: $2.000 per 1M input tokens and $10.000 per 1M output tokens, with cached input at 95% off the input rate ($0.100). Its standard prices are $4.000 and $20.000, with cached input at $0.200. Google has not announced when Carbon's introductory period ends. EvoLink's current rates follow the introductory price; when it ends, our usual discount applies to Google's standard price.

Does Gemini 4 Carbon think by default, and how do I control it?

Yes. Gemini 4 Carbon uses thinking tokens even on short prompts, and they are billed at the output rate. Choose low, medium, or high for each request: reasoning_effort in Chat Completions, or thinkingLevel in thinkingConfig on the native API. Low is the lightest level.

What is Gemini 4 Carbon's context window and output limit?

Google has not published Gemini 4 Carbon’s context window or maximum response length. The native API accepts maxOutputTokens up to 1,048,576 as a request budget, which is not a guarantee of generated length. Output includes thinking tokens. Set a budget for each task and test long inputs on your own documents.

Which protocols work with Gemini 4 Carbon?

The OpenAI-compatible Chat Completions endpoint and the Gemini native API, both on direct.evolink.ai with the same EvoLink API key and the same model ID. Both stream; on the native API, call streamGenerateContent with alt=sse. Both Chat Completions and the Gemini native API stream Carbon's output, and the final chunk carries usage. Show text as it arrives and still reconcile the billed tokens for every request.

How is Gemini 4 Carbon different from Gemini 4 Argon on EvoLink?

Carbon and Argon use the same EvoLink API key, endpoints, request format and prices, so you can switch by changing the model ID. Evaluate Carbon on the same coding or reasoning tasks, with the same tools and acceptance checks. Compare accepted results, thinking tokens, retries, elapsed time and total task cost. Keep Argon where Carbon does not improve the result or justify the extra work.

Does Gemini 4 Carbon accept images or call tools on EvoLink?

This page lists what we have verified for Gemini 4 Carbon on EvoLink: text input and output, thinking levels, and streaming on both protocols. Image, audio, and video input, function calling, structured output, and Google Search will be added here after we test them on Carbon.