Gemini 4 Carbon API
gemini-4-carbonGemini 4 Carbon is Google's deeper-reasoning Gemini 4 model. Google gave cyber defenders access first and, about two weeks later, on 10 November 2026, opened the API to every paid Gemini API developer. On APIMODELS it costs $1.10 per 1M input tokens and $5.50 per 1M output tokens through 30 November 2026, 45% below Google's $2 / $10 list, with cached input at $0.055 (Google: $0.10). Google built Carbon for harder problems — multi-step debugging and open-ended research — and says it spends more tokens to explore thoroughly, reason deeply and run verification loops before it commits to an answer. Google is just as clear that Gemini 4 Argon remains the better choice for most software-engineering and knowledge work, so treat Carbon as the model you escalate to when a problem resists Argon, not as a default replacement. The Gemini 4 Carbon benchmarks Google published: first place on CursorBench 4.0 and on Perplexity's WANDR; 92% of 375 tasks passed on an agentic honesty benchmark, against 70% for Argon; and in Gray Swan's automated red-teaming, cyberattack queries succeeded 2.17% of the time against Carbon versus 9.86% against Argon (lower is safer). These are Google's own figures, so run your own prompts before moving a production workload. Specifications: a 1M-token context window shared by input and output, up to 65,536 output tokens per response (Google's separate long-decode mode goes to 1M, but we have not verified it through APIMODELS), text, image, video, audio and PDF input, text output, and adjustable thinking levels. Thinking tokens bill at the output rate, and because Carbon thinks longer, the same task usually costs more on Carbon than on Argon even though the per-token prices are identical — estimate cost from total output, not from the visible answer. Latency follows the same pattern: in our first-week tests a one-word reply took about 4–6 seconds, and real tasks can run for minutes, so stream long requests. Caching is Google's implicit caching only: whether a repeated prompt prefix hits is decided on Google's side and is not guaranteed; cached tokens bill at $0.055 per 1M ($0.11 from 1 December), and explicit cache objects are not supported, so plan costs as if nothing is cached. Google offers no free tier for Carbon. On pricing: Google lists Carbon at $2 / $10 (cached $0.10) through 30 November 2026 and at $4 / $20 (cached $0.20) from 1 December 2026. APIMODELS follows the same day: from 00:00 UTC on 1 December 2026, Carbon bills at $2.20 per 1M input and $11.00 per 1M output, with cached input at $0.11 — still 45% below Google's list. The new rate applies automatically; no code change is needed. Call it with model gemini-4-carbon on the native Gemini endpoint or the OpenAI-compatible /v1/chat/completions, one API key for both; requests switch to another upstream channel automatically if one fails, and failed requests are not charged. When not to use it: for everyday coding and knowledge work, Gemini 4 Argon at the same per-token price is usually the better fit, and for short, high-volume calls Gemini 3.8 Flash ($0.450 / $2.250) is the more affordable Google model here.
API Documentation
View complete API reference with all parameters and examples.
Full API Documentation
View complete API reference with streaming, thinking, and more.
Pricing
Billing: Cost = (input_tokens * input_price + output_tokens * output_price) / 1,000,000
45% Below Google List
$1.10 / $5.50 per 1M tokens through 30 November (Google: $2 / $10); $2.20 / $11.00 from 1 December (Google: $4 / $20)
Gemini 4 Carbon Benchmarks
Google: #1 on CursorBench 4.0 and Perplexity's WANDR; agentic honesty 92% of 375 tasks (Argon 70%); Gray Swan cyberattack success 2.17% (Argon 9.86%)
When to Pick Carbon over Argon
Google: Carbon for multi-step debugging and open-ended research; Argon stays the better fit for most coding and knowledge work
1M Context, 65,536-Token Output
1M tokens shared by input and output; up to 65,536 output tokens per response; text, image, video, audio and PDF input; adjustable thinking levels
More Thinking per Task
Same per-token price as Argon, but more thinking tokens billed as output, so a task usually costs more; implicit caching only, cached reads $0.055 per 1M ($0.11 from 1 December)
New Rate on 1 December, Automatically
From 00:00 UTC on 1 December 2026, in step with Google: $2.20 / $11.00 per 1M, cached $0.11; no code change
About Gemini 4 Carbon
Gemini 4 Carbon is a Large Language Model API provided by Google. Gemini 4 Carbon is Google's deeper-reasoning Gemini 4 model. Google gave cyber defenders access first and, about two weeks later, on 10 November 2026, opened the API to every paid Gemini API developer. On APIMODELS it costs $1.10 per 1M input tokens and $5.50 per 1M output tokens through 30 November 2026, 45% below Google's $2 / $10 list, with cached input at $0.055 (Google: $0.10). Google built Carbon for harder problems — multi-step debugging and open-ended research — and says it spends more tokens to explore thoroughly, reason deeply and run verification loops before it commits to an answer. Google is just as clear that Gemini 4 Argon remains the better choice for most software-engineering and knowledge work, so treat Carbon as the model you escalate to when a problem resists Argon, not as a default replacement. The Gemini 4 Carbon benchmarks Google published: first place on CursorBench 4.0 and on Perplexity's WANDR; 92% of 375 tasks passed on an agentic honesty benchmark, against 70% for Argon; and in Gray Swan's automated red-teaming, cyberattack queries succeeded 2.17% of the time against Carbon versus 9.86% against Argon (lower is safer). These are Google's own figures, so run your own prompts before moving a production workload. Specifications: a 1M-token context window shared by input and output, up to 65,536 output tokens per response (Google's separate long-decode mode goes to 1M, but we have not verified it through APIMODELS), text, image, video, audio and PDF input, text output, and adjustable thinking levels. Thinking tokens bill at the output rate, and because Carbon thinks longer, the same task usually costs more on Carbon than on Argon even though the per-token prices are identical — estimate cost from total output, not from the visible answer. Latency follows the same pattern: in our first-week tests a one-word reply took about 4–6 seconds, and real tasks can run for minutes, so stream long requests. Caching is Google's implicit caching only: whether a repeated prompt prefix hits is decided on Google's side and is not guaranteed; cached tokens bill at $0.055 per 1M ($0.11 from 1 December), and explicit cache objects are not supported, so plan costs as if nothing is cached. Google offers no free tier for Carbon. On pricing: Google lists Carbon at $2 / $10 (cached $0.10) through 30 November 2026 and at $4 / $20 (cached $0.20) from 1 December 2026. APIMODELS follows the same day: from 00:00 UTC on 1 December 2026, Carbon bills at $2.20 per 1M input and $11.00 per 1M output, with cached input at $0.11 — still 45% below Google's list. The new rate applies automatically; no code change is needed. Call it with model gemini-4-carbon on the native Gemini endpoint or the OpenAI-compatible /v1/chat/completions, one API key for both; requests switch to another upstream channel automatically if one fails, and failed requests are not charged. When not to use it: for everyday coding and knowledge work, Gemini 4 Argon at the same per-token price is usually the better fit, and for short, high-volume calls Gemini 3.8 Flash ($0.450 / $2.250) is the more affordable Google model here. Through APIMODELS platform, you can access this model via a unified API with transparent pay-as-you-go pricing. Current pricing: Input: $1.10, Output: $5.50 per 1M tokens.
Key Features
- 45% Below Google List -- $1.10 / $5.50 per 1M tokens through 30 November (Google: $2 / $10); $2.20 / $11.00 from 1 December (Google: $4 / $20)
- Gemini 4 Carbon Benchmarks -- Google: #1 on CursorBench 4.0 and Perplexity's WANDR; agentic honesty 92% of 375 tasks (Argon 70%); Gray Swan cyberattack success 2.17% (Argon 9.86%)
- When to Pick Carbon over Argon -- Google: Carbon for multi-step debugging and open-ended research; Argon stays the better fit for most coding and knowledge work
- 1M Context, 65,536-Token Output -- 1M tokens shared by input and output; up to 65,536 output tokens per response; text, image, video, audio and PDF input; adjustable thinking levels
- More Thinking per Task -- Same per-token price as Argon, but more thinking tokens billed as output, so a task usually costs more; implicit caching only, cached reads $0.055 per 1M ($0.11 from 1 December)
- New Rate on 1 December, Automatically -- From 00:00 UTC on 1 December 2026, in step with Google: $2.20 / $11.00 per 1M, cached $0.11; no code change
Use Cases
Chatbot & Customer Support
Build intelligent conversational systems to automatically answer user queries and improve service efficiency.
Content Generation
Automatically write articles, emails, ad copy, and other text content to boost productivity.
Code Assistant
Assist with code writing, debugging, and code review to accelerate software development.
Data Analysis
Understand and analyze unstructured data, extract key insights, and generate summary reports.
Why APIMODELS
- Unified API -- One API key to access all models, no need to register on multiple platforms
- Transparent Pricing -- Pay-as-you-go with per-call pricing published on every model page, ideal for indie developers and startups
- Instant Access -- Start right after signup; pay by card, PayPal, Alipay or USDT
- Full Documentation -- Detailed API docs with code examples in cURL, Python, and Node.js
Frequently Asked Questions
What is the difference between APIMODELS and the official Google API?
APIMODELS offers the same Gemini 4 Carbon model through our aggregation platform. We provide a unified API interface so you do not need separate accounts for each provider - one API key to access all models.
What is Gemini 4 Carbon API pricing on apimodels, and what changes on 1 December?
Through 30 November 2026 it is $1.10 per 1M input tokens and $5.50 per 1M output, with cached input at $0.055 — 45% below Google's list of $2 / $10 and $0.10 cached, and exactly the same per-token rates as Gemini 4 Argon. Thinking tokens bill at the output rate, as they do at Google, and there is no long-context surcharge: Google lists one rate for the whole 1M window. From 1 December 2026 Google lists Carbon at $4 / $20, with cached input at $0.20, and APIMODELS moves the same day: from 00:00 UTC on 1 December 2026, Carbon bills at $2.20 per 1M input and $11.00 per 1M output, with cached input at $0.11 — still 45% below Google's list. Until then every call bills at $1.10 / $5.50. The switch is automatic; there is nothing to change in your code. You pay only for the tokens used, failed requests are not charged, and there is no subscription.
Gemini 4 Carbon vs Gemini 4 Argon: when should I use Carbon?
Google's own answer is that Argon remains the better choice for most software-engineering and knowledge work, and Carbon is for harder problems: multi-step debugging, open-ended research, and tasks where the first plausible answer is often wrong. Carbon gets there by spending more tokens — it explores more options, reasons longer and runs verification loops before it commits. In practice, keep Argon as your default and send a task to Carbon when Argon stalls or when a wrong answer is expensive. On APIMODELS both bill at the same per-token rates ($1.10 / $5.50 per 1M through 30 November, $2.20 / $11.00 from 1 December), but Carbon usually uses more thinking tokens, so the same job usually costs more and takes longer. Switching is one field — model: gemini-4-carbon — on the same key and endpoints. For short, high-volume calls, Gemini 3.8 Flash ($0.450 / $2.250) is a better fit than either.
Is Gemini 4 Carbon free?
No. Google offers no free tier for Gemini 4 Carbon — it is only on the paid tier of the Gemini API — and it bills per token here too. New APIMODELS accounts get $0.10 of trial credit (one new account per IP address every 48 hours). For scale: through 30 November, a request with 10,000 input tokens and 4,000 output tokens (thinking included) costs $0.011 + $0.022 = $0.033, so the trial covers a few short test calls; from 1 December the same request costs $0.022 + $0.044 = $0.066. After that you top up from $10 and pay only for what you use, with no subscription.
What are the Gemini 4 Carbon benchmarks?
From Google's launch post: first place on CursorBench 4.0 and on Perplexity's WANDR; on an agentic honesty benchmark of 375 tasks, Carbon passed 92% against 70% for Argon; and in Gray Swan's automated red-teaming, cyberattack queries succeeded 2.17% of the time against Carbon versus 9.86% against Argon, where lower is safer. These are Google's own numbers rather than independent runs, so treat them as a starting point and compare on your own prompts — switching between Carbon and Argon here is a change to the model field.
How do I call the Gemini 4 Carbon API?
Set the model to gemini-4-carbon and authenticate with your apimodels key. In the native Gemini format, POST https://api.apimodels.app/v1beta/models/gemini-4-carbon:generateContent (:streamGenerateContent for streaming) with a contents array; thinkingConfig.thinkingLevel sets how much it thinks. In the OpenAI format, POST https://api.apimodels.app/v1/chat/completions with the model set to gemini-4-carbon, so the OpenAI SDK works once base_url is https://api.apimodels.app/v1. Stream anything non-trivial: Carbon thinks before it writes, in our first-week tests even a one-word reply took about 4–6 seconds, and real tasks can run for minutes, long enough for a non-streaming request to hit a client or proxy timeout. You do not need a Google Cloud project or billing account, it works from mainland China, and if one upstream channel fails the request switches to another automatically.
Why does the same task cost more on Gemini 4 Carbon than on Argon?
Because the per-token prices are identical but the token counts are not. Thinking tokens bill at the output rate, and Carbon is designed to think more — exploring, reasoning and checking its own work — so a task usually produces more output tokens on Carbon than on Argon. Estimate cost from candidatesTokenCount plus thoughtsTokenCount in usageMetadata, not from the length of the text you see, and set maxOutputTokens (max_tokens in the OpenAI format) on open-ended tasks. Where the task allows it, a lower thinkingLevel trims the thinking. On caching, Carbon supports Google's implicit caching only: whether a repeated prefix hits the cache is decided on Google's side and is not guaranteed, and any cached tokens bill at $0.055 per 1M ($0.11 from 1 December). Explicit cache objects cannot be created, so plan costs as if nothing is cached.
What are the Gemini 4 Carbon specs and limits?
A 1M-token context window shared by input and output, and up to 65,536 output tokens per response. Google also offers a separate long-decode mode that goes to 1M output tokens, but we have not verified it through APIMODELS. Input can be text, image, video, audio or PDF; output is text; thinking levels are adjustable. At Google it is available only on the paid tier of the Gemini API, with no free tier; here it is pay-as-you-go from the same balance as every other model. Output, thinking included, bills at $5.50 per 1M tokens ($11.00 from 1 December), so cap maxOutputTokens when you do not need the full length. Long tasks can take minutes, so stream them; failed requests are not charged.
What can Gemini 4 Carbon do?
Gemini 4 Carbon supports: 45% Below Official, Deep Reasoning, 1M Context, CursorBench 4.0 #1. See the APIMODELS docs for full parameters and call examples.
Can I use the Gemini 4 Carbon API from any country, including mainland China?
Yes. APIMODELS does not restrict access by country: the site and the API answer requests from anywhere, and apimodels.app and api.apimodels.app are reachable directly from mainland China without a proxy. You call Gemini 4 Carbon with one APIMODELS key, so you do not need a separate account with the model's developer.
What payment methods are supported? Can I pay with PayPal or crypto?
Yes to both. You can top up with a card (Visa, Mastercard and others, through Stripe), PayPal, Alipay, or USDT on TRON (TRC20) or BNB Smart Chain (BEP20). Card, PayPal and Alipay top-ups are credited immediately; USDT is credited once the transfer confirms on-chain. Auto top-up is optional, and a $50 top-up adds a $1 bonus ($3 on a $100 top-up). The balance is in USD and billed per use, with no subscription.
Related models
Made and tested with Gemini 4 Carbon
See all 223 →Prompts shared by their authors — copy and adapt them. Each one credits its author and links back to the original post.
Airplane cutaway in 3D, Carbon vs Opus 5.5
No prompt published. A side-by-side 3D test of Carbon's Arena version against Opus 5.5. The author says Carbon is seriously good at 3D and t…
by @Mr_Salio
Low-poly running giraffe, Opus 5.5 vs Carbon
No prompt published. Opus 5.5 and Carbon side by side, offered as proof that Carbon is not just tuned for benchmarks.
by @Haleeeemahh
Rainbow jelly you can cut, nail and torch
Only the creator's request is public, not a full prompt: "rainbow jelly you can cut, nail and torch."
by @0xRhinoX
Need prompts?
We curate copy-ready prompt libraries — every entry shows its full text and a sample result, ready to adapt.
Related reading
How to get access, regional availability, and how this model compares with its alternatives.