Updated
What is the Gemini 4 Neon API?
The Gemini 4 Neon API is live on UnificAlly as google/gemini-4-neon, from $0.90/1M tokens of input and $4.50/1M tokens of output. Gemini 4 Neon is Google's Gemini 4 model for complex coding, analytical reasoning, and multi-step agent work, released November 18, 2026. Google places it between Gemini 4 Hydrogen and Gemini 4 Carbon: close to Carbon on coding and reasoning, at much higher speed and a lower official price. It has a 1M-token context window and takes text and image input. It runs on UnificAlly as google/gemini-4-neon, with two reasoning effort levels (low and high).
Gemini 4 Neon API specs
| Model ID | google/gemini-4-neon |
| Endpoint | POST https://direct-api.unifically.com/v1/chat/completions |
| Request types | Chat, streaming, image input, tool calls, JSON mode |
| Input | Text and images |
| Output | Text |
| Context window | 1M tokens |
| Reasoning effort | low (default), high |
| Image input | An image_url content part with a direct https link or a data:image/...;base64 URI |
| Input rate | $0.90/1M tokens |
| Output rate | $4.50/1M tokens |
| Cache read rate | $0.045/1M tokens |
The Gemini 4 Neon API bills per token, with no subscription. Thinking tokens are billed as output tokens. The Gemini API pricing guide covers the rates for every Gemini model.
How to use the Gemini 4 Neon API
- Create an API key on the API keys page.
- Send a POST request to
/v1/chat/completionsonhttps://direct-api.unifically.comwith the model ID and amessagesarray. - Read the reply from the stream when
streamistrue, or fromchoices[0].message.contentin the same response.
curl -X POST https://direct-api.unifically.com/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "google/gemini-4-neon",
"reasoning_effort": "high",
"stream": true,
"messages": [
{"role": "user", "content": "Compare these three vendor quotes, flag the hidden costs, and recommend one in a short table."}
]
}'
Tokens arrive as they generate. The stream is sent as server-sent events and ends with data: [DONE]. The first token usually takes 10 to 30 seconds, and longer at high, so stream the reply and give your client timeout some headroom. Send requests to direct-api.unifically.com so long replies are not cut short by a proxy timeout.
Without stream, the response comes back synchronously with the completion and the token usage:
{
"id": "7c1d4e2a-9b3f-4a8e-8d2c-6f0e1b9a5c34",
"object": "chat.completion",
"created": 1710000000,
"model": "google/gemini-4-neon",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "Vendor B is cheapest once the setup fee and the annual support charge are counted..."
},
"finish_reason": "stop"
}
],
"usage": {
"prompt_tokens": 31,
"completion_tokens": 72,
"total_tokens": 103
}
}
Gemini 4 Neon API parameters
| Parameter | Type | Default | What it does |
|---|---|---|---|
model | string | required | Model identifier, google/gemini-4-neon |
messages | array | required | Non-empty array of message objects with role and content (a string or an array of content parts). Roles: system, developer, user, assistant, tool |
stream | boolean | false | If true, the response is streamed using server-sent events |
reasoning_effort | string | low | How much the model thinks before answering: low or high. minimal maps to low; medium, xhigh and max map to high |
tools | array | none | List of tools the model may call. Each tool needs a type and a function definition |
tool_choice | string or object | none | Controls which tool, if any, is called: none, auto, required, or a specific tool |
response_format | object | none | Set { "type": "json_object" } to get a JSON reply |
The Gemini 4 Neon API documentation has the full reference.
Image input request
Send an image in an image_url content part alongside your text, either as a direct https link or inline as a base64 data URI. The link must point straight at the image file, with no redirect. Images are counted as input tokens.
curl -X POST https://direct-api.unifically.com/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "google/gemini-4-neon",
"messages": [
{"role": "user", "content": [
{"type": "text", "text": "Read every number in this chart and return them as a JSON table."},
{"type": "image_url", "image_url": {"url": "https://example.com/chart.png"}}
]}
]
}'
Gemini 4 Neon key features
#3 on the Artificial Analysis Intelligence Index
Google reports 61 on the Artificial Analysis Intelligence Index at its top thinking level, third overall and ahead of comparable models, just behind larger and slower ones. Gemini 4 Carbon scores 60 and Gemini 4 Hydrogen 54.
#1 for coding agents
It ranks first on the Artificial Analysis Coding Agent Index at 58.6 and scores 65.2% on Terminal-Bench 4.1, against 57% for Gemini 4 Hydrogen.
Long-running agent work
It debuts #2 on Arena's Agent Arena Index at 1336 across more than 4,200 real agent sessions, and takes #1 on Zapier AutomationBench (56.3%) and Vending-Bench 2, a long business simulation.
Legal, tax, and science reasoning
Vals AI ranks it #1 on Legal Research Bench and Tax Agent Bench. Google reports 71.3% on HLE-Diamond, 84.0% on RiemannBench, and 70.5% on Terminal-Bench Science.
What Gemini 4 Neon is best for
Coding agents
#1 on the Artificial Analysis Coding Agent Index and 65.2% on Terminal-Bench 4.1.
Business automation
#1 on Zapier AutomationBench at 56.3% and on Vending-Bench 2, a long business simulation.
Legal and tax research
#1 on the Vals AI Legal Research Bench and Tax Agent Bench.
Reports from dense material
Turns spreadsheets, PDFs, and long briefs into structured summaries and tables.
Charts and screenshots
73.5% on Chartography for reading dense professional charts, with image input on every request.
Whole-codebase context
A 1M-token window keeps a repository and a long tool history in one call.
Gemini 4 Neon use cases
Run it as the main model in a coding agent that reads the repo, runs the tests, and fixes the failure. Hand it a stack of invoices, statements, or receipts and ask which charges do not match. Build legal and tax research pipelines that pull the relevant rules and cite them. Send it a chart or a screenshot and get the numbers back as JSON. Keep low for routine steps and switch a single request to high when the task needs deeper thought.
Gemini 4 Neon limitations
It is slow to start. The first token usually takes 10 to 30 seconds, and longer at high, so it is a poor fit for chat that has to feel instant. Stream every request.
Only two thinking levels are offered here. low is the default, and medium, xhigh and max all run as high. The benchmark scores Google quotes were measured at its top level.
Images go in as direct https links or inline base64 data URIs, and a link that redirects is refused. Video, audio, and file input are not available on this model yet.
Gemini 4 Neon vs Gemini 4 Carbon
Neon matches or passes Carbon on most coding and reasoning tests while running faster. It scores 61 on the Artificial Analysis Intelligence Index against 60 for Carbon. Google's official per-token price for Neon is also lower than Carbon's from December 1, 2026, when Carbon's launch pricing ends. Carbon is still Google's pick for the longest, hardest projects.
Gemini 4 Neon vs Gemini 4 Hydrogen
Neon is the stronger reasoner and Hydrogen is the cheaper worker. Neon leads the Artificial Analysis Intelligence Index 61 to 54 and Terminal-Bench 4.1 65.2% to 57%. Hydrogen answers sooner and costs half as much per token, so it suits simple, high-volume steps.
When to use Gemini 4 Neon
Use Gemini 4 Neon when the task needs real reasoning: coding agents, research, and analysis where the answer has to be right. Drop to Gemini 4 Hydrogen for quick, high-volume steps. Step up to Gemini 4 Carbon for the longest projects.
FAQs
People also ask about Gemini 4 Neon
google/gemini-4-neon, called through the OpenAI-compatible POST /v1/chat/completions endpoint with your UnificAlly API key.
Create a key on the API keys page and send it as Authorization: Bearer <key> to https://direct-api.unifically.com/v1/chat/completions. The endpoint speaks the OpenAI chat format, so existing OpenAI client code only needs that base URL and the model name google/gemini-4-neon.
Pass google/gemini-4-neon, a messages array (images as image_url parts) and, if you want deeper thinking, reasoning_effort: "high". Set "stream": true so tokens arrive as server-sent events while the model works; without it, read choices[0].message.content from the synchronous response.
Text and image input, with text output. Send images as direct https links or inline as base64 data URIs. The model has a 1M-token context window.
Two. reasoning_effort takes low, the default, and high. minimal is treated as low, and medium, xhigh and max are treated as high. Thinking tokens are billed as output tokens.
Google reports 61 on the Artificial Analysis Intelligence Index at its top thinking level, #3 overall, and #1 on the Artificial Analysis Coding Agent Index at 58.6. It scores 65.2% on Terminal-Bench 4.1 and 71.3% on HLE-Diamond, and debuts #2 on Arena's Agent Arena Index at 1336.
Yes, for hard coding and analysis. Neon scores 61 on the Artificial Analysis Intelligence Index against 54 for Hydrogen, and 65.2% on Terminal-Bench 4.1 against 57%. Hydrogen stays the cheaper, faster pick for simple high-volume steps.
The first token usually arrives after 10 to 30 seconds, longer at high. Stream the response and give your client timeout some headroom.
Google released it on November 18, 2026. It sits between Gemini 4 Hydrogen and Gemini 4 Carbon in the Gemini 4 family and replaces Gemini 3.1 Pro in the Gemini app.
Gemini 4 Neon guides and comparisons
Review pricing, limits, and tested outputs before running this model.
Models similar to Gemini 4 Neon
Browse all modelsGemini 4 Carbon API
60 on the Artificial Analysis Intelligence Index, #1 on the Vals Finance Index (67.1%), 70.5% Vals Index, 1M context.
- Text to Text
Gemini 4 Hydrogen API
54 on the Artificial Analysis Intelligence Index at xhigh, #1 on Tax Agent Bench (52.3%), 2.5x faster than Gemini 3.8 Flash, 1M context.
- Text to Text
