Gemini 4 Neon by Google (Multimodal)
Gemini 4 Neon by Google is a multimodal reasoning model for coding, analytics, and document work. It supports Google Search and Maps grounding and a 1,048,576-token window.
## Overview
Gemini 4 Neon is a Google DeepMind model tuned for complex coding, knowledge work, and analytical tasks. It sits between Gemini 4 Hydrogen and Gemini 4 Carbon, giving frontier-level reasoning with fast end-to-end task completion. It reads text plus images, audio, video, and PDFs, then produces structured, task-ready writing or code. Grounding with Google Search or Google Maps helps reduce hallucinations and adds source metadata for citations.
## What you can build
- Repo-level coding help, including multi-file refactors, debugging, and patch drafting
- Long document workflows like contract summaries, policy comparisons, and due diligence notes
- Business analysis write-ups based on tables, charts, and attached files
- Research assistants that ground answers in Google Search results with citation metadata
- Local planning assistants that use Google Maps place data for location-aware answers
- Multimodal Q&A, such as explaining a chart screenshot, a slide, or a recorded meeting
## Inputs
- A task prompt in plain text, like a question, instruction, or brief
- Optional chat-style context as role-based messages for multi-turn conversations
- Optional system instructions that set rules for tone, format, and boundaries
- Optional file attachments: images, PDFs, audio, and video
- Optional grounding with Google Search for fresh web information, or Google Maps for places and reviews
- Optional thinking level: low (default on this listing), medium, high, or xhigh
- Optional response length budget, up to 65,536 tokens per response
- Optional user and session IDs so chat history stays separated per user and conversation
## Outputs
- A text response with explanations, step-by-step plans, code, or formatted deliverables such as checklists and tables.
- With Google Search or Google Maps grounding, structured grounding metadata that maps parts of the answer to its sources, which apps can show as inline citations.
## Recommended settings
- Keep low for everyday questions and fast chat.
- Use high for complex coding, analysis, and multi-step tasks.
- Reserve xhigh for hard engineering or scientific problems where high falls short, since it uses more tokens and takes longer.
## Limitations
- Input, thinking, and output share one 1,048,576-token window, and a single response is capped at 65,536 tokens.
- Thinking tokens count toward the response budget and are billed as output.
- Temperature, top-p, and top-k are ignored. Use the thinking level and clear formatting instructions instead.
- Google Search and Google Maps grounding can't be used together. When Maps is on, web search is turned off.
- Ungrounded answers can be outdated. Turn on Google Search for current information.
- Scanned PDFs, handwriting, small text, and dense layouts increase the risk of misreading.
## Safety & compliance
- Follow Google's Gemini usage policies.
- Don't upload sensitive personal data without a clear legal basis and user consent.
- Treat generated content as a draft and verify critical claims, especially ungrounded ones.
- For grounded answers, review the cited sources before relying on specific statements.
API quick start
Run gemini 4 neon with a single API call.
{
"prompt": "Plan a 3-day trip to Lisbon for two peopl…",
"inputAll": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}curl -X POST "https://api.wiro.ai/v1/Run/google/gemini-4-neon" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "Plan a 3-day trip to Lisbon for two peopl…",
"inputAll": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}
JSONOther models you might like
Explore similar llm & chat models.



