Gemini 4 Carbon API
Gemini 4 Carbon is Google's strongest frontier model, built on Gemini 4 Argon for long-horizon agentic workflows, cybersecurity, scientific reasoning, and complex knowledge work.
Gemini 4 Carbon is Google's frontier model released on October 28, 2026, built on top of Gemini 4 Argon. It accepts text, image, audio, and video input and produces text output, with a combined input and output context of 1 million tokens. Its knowledge cutoff is March 2026. Configurable thinking levels let users balance quality, latency, and cost, and the model uses more thinking tokens than Gemini 4 Argon to explore hypotheses, reason deeply, and run extended verification loops.
The model targets long-horizon agentic work across complex knowledge work, software engineering, cybersecurity, and scientific reasoning. It supports engineering tasks requiring multi-agent collaboration, everyday debugging, large-scale codebase migrations, and algorithm design, as well as high-precision knowledge work, artifact generation, and data analysis. For science, it can translate security goals into mathematical theorems and autonomously discover remote homologs across petabase-scale sequence archives, increasing homolog retrieval by 40-fold on average for low-homolog protein families.
On safety and reliability, Google reports that high-severity misalignment alerts dropped by almost 80% in live coding environments, and that Gemini 4 Carbon outperforms Gemini 4 Argon on multilingual and multimodal content safety with a 3.50 percentage point improvement on text-to-text safety evaluations. Known limitations include potential hallucinations, occasional slowness or timeout issues, and higher token usage to maximize performance, especially at higher thinking levels. It was initially released to a subset of cybersecurity defenders through the Fairwind Program.
Updated:
MODEL ID
gemini-4-carbon- Context
- 1.049M
- Max output
- 943.718K
- Recommended API
- messages
- Input
- audio / file / image / text / video
- Output
- text
Highlights
- 1 million token combined context
- Configurable thinking levels
- Text, image, audio, and video input
- Built on Gemini 4 Argon
- March 2026 knowledge cutoff
- Leads the full vulnerability lifecycle
Providers
Best priceCompare API prices across providers. Select a provider to view all charges and conditions.
| Uptime | |||||
|---|---|---|---|---|---|
Historical observations are not a live health check. Missing data does not mean an outage.
What $10 buys
- Input
- 2.5M tokens
- Output
- 500K tokens
Estimated from current input and output prices, before discounts.
Benchmarks
- 59
Artificial Analysis Intelligence Index v5.0
Top spot; Claude Opus 5.5 scored 58
- 88.2%
SEC-bench Pro
Vulnerability finding
- 54.8%
ExploitGym
Vulnerability exploitation
- 71.0%
CWE-bench
Vulnerability patching
- 0.0%
Terminal-Bench 4.0
Successful cheating rate
- 70.45%
Vals Index
Rank 3 of 36 models
- 55.77%
Legal Research Bench
Rank 1
- #1
General Financial Intelligence benchmark
Top out of 23 frontier model configurations; Balyasny Asset Management
Scores as published by the model provider.
Reasoning
- Default effort
- medium
- Always enabled
- Yes
- Available effort levels
- high / medium / low
Quick Start
Drop-in code to call this model with Felo API Platform compatible APIs. Start in the Playground when you want to test prompts before writing code.
Get your API key
Create a key from your Felo API Platform account and set it as an environment variable.
export FELO_API_KEY="YOUR_API_KEY"Make your first request
Use this model ID with the Felo API Platform endpoint. Optional request headers such as X-Request-Id can help troubleshooting.
gemini-4-carbonThis model is reasoning-capable. If your selected protocol and model route expose reasoning controls, preserve the returned reasoning metadata when continuing a conversation.
const response = await fetch("https://openapi.felo.ai/api/v1/messages", {
method: 'POST',
headers: {
Authorization: `Bearer ${process.env.FELO_API_KEY}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
"model": "gemini-4-carbon",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello"
}
]
}),
});
if (!response.ok) throw new Error(await response.text());
console.log(await response.json());Using third-party SDKs
Use compatible SDKs by setting baseURL to https://openapi.felo.ai/api/v1. Use the Anthropic-compatible base URL for Claude-style clients. https://openapi.felo.ai/api.
Enable streaming
Add stream: true to receive responses as server-sent events.
curl -N https://openapi.felo.ai/api/v1/messages \
-H "Authorization: Bearer $FELO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-4-carbon",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello"
}
],
"stream": true
}'Endpoint
Felo API Platform exposes compatible model-calling surfaces. Pick the surface your app or agent client already supports.
Messages
Anthropic-compatible messages for Claude-style clients.
POST
https://openapi.felo.ai/api/v1/messagesParameters
modelstringgemini-4-carbonPublic model ID to call.max_tokensinteger—Maximum generated tokens.messagesarray—Conversation messages for Chat Completions.seed———stop———toolsarray—Tool calling definitions following compatible schema.top_p———reasoning———temperaturenumber—Controls response variety when supported.tool_choicestring or object—Controls whether a tool should be called.response_formatobject—Request structured output when supported.reasoning_effort———include_reasoning———structured_outputs———Frequently asked questions
What is Gemini 4 Carbon?
Gemini 4 Carbon is Google's frontier model released on October 28, 2026, built on top of Gemini 4 Argon. It accepts text, image, audio, and video input and produces text output, with a combined input and output context of 1 million tokens. Its knowledge cutoff is March 2026. Configurable thinking levels let users balance quality, latency, and cost, and the model uses more thinking tokens than Gemini 4 Argon to explore hypotheses, reason deeply, and run extended verification loops. The model targets long-horizon agentic work across complex knowledge work, software engineering, cybersecurity, and scientific reasoning. It supports engineering tasks requiring multi-agent collaboration, everyday debugging, large-scale codebase migrations, and algorithm design, as well as high-precision knowledge work, artifact generation, and data analysis. For science, it can translate security goals into mathematical theorems and autonomously discover remote homologs across petabase-scale sequence archives, increasing homolog retrieval by 40-fold on average for low-homolog protein families. On safety and reliability, Google reports that high-severity misalignment alerts dropped by almost 80% in live coding environments, and that Gemini 4 Carbon outperforms Gemini 4 Argon on multilingual and multimodal content safety with a 3.50 percentage point improvement on text-to-text safety evaluations. Known limitations include potential hallucinations, occasional slowness or timeout issues, and higher token usage to maximize performance, especially at higher thinking levels. It was initially released to a subset of cybersecurity defenders through the Fairwind Program.
How is usage billed?
Your bill depends on the provider you use, its prices and your usage. Longer inputs or other conditions may have different prices.
What input types does Gemini 4 Carbon support?
It accepts text, image, audio, and video input and returns text output.
How large is its context window?
It supports 1 million tokens combined input and output.
What is Gemini 4 Carbon best suited for?
It is built for long-horizon agentic workflows across complex knowledge work, software engineering, cybersecurity, and scientific reasoning, including multi-agent engineering tasks and vulnerability discovery and patching.
How does it differ from Gemini 4 Argon?
It is built on top of Gemini 4 Argon and uses more thinking tokens to explore hypotheses, reason deeply, and run extended verification loops. It also outperforms Gemini 4 Argon on multilingual and multimodal content safety, with a 3.50 percentage point improvement on text-to-text safety evaluations.
What are its known limitations?
Potential hallucinations, occasional slowness or timeout issues, and higher token usage to maximize performance, especially at higher thinking levels.
Related models
Gemini 4 Carbon (batch)
Googlegemini-4-carbon:batchGemini 4 Carbon is Google's multimodal frontier model for long-horizon agentic work, with 1 million token context and adjustable reasoning effort.
Context
1.049M
Max output
943.718K
Input tokens
$2 / 1M tokens
Output tokens
$10 / 1M tokens
Gemini 4 Argon
Googlegemini-4-argonGoogle's Gemini 4 Argon is a frontier model with a 1,048,576-token context window, multimodal input support, and top scores on 14 of 18 internal benchmarks.
Context
1.049M
Max output
943.718K
Input tokens
$4 / 1M tokens
Output tokens
$20 / 1M tokens
Gemini 3.8 Flash
Googlegemini-3.8-flashGemini 3.8 Flash is Google's most intelligent Flash model, a reasoning model with 1,000,000-token context and multimodal input for long-horizon agents and enterprise workflows.
Context
1.049M
Max output
65.536K
Input tokens
$1.5 / 1M tokens
Output tokens
$7.5 / 1M tokens
Gemma 4 26B A4B (free)
Googlegemma-4-26b-a4b-it:freeGemma 4 26B A4B is an open-weight multimodal Mixture-of-Experts model from Google DeepMind with 26B total and 4B active parameters, a 256K token context, and support for text, image, and video input.
Context
262.144K
Max output
32.768K
Input tokens
$0 / 1M tokens
Output tokens
$0 / 1M tokens
Gemini 2.5 Pro
Googlegemini-2.5-proGoogle DeepMind's Gemini 2.5 Pro is a hybrid-reasoning, multimodal model with a 1 million token context window, built for advanced coding, agentic applications, and video understanding.
Context
1.049M
Max output
65.536K
Input tokens · Base rate
$1.25 / 1M tokens
Output tokens · Base rate
$10 / 1M tokens