Gemini 4 Neon — frontier engineering and deep reasoning
Gemini 4 Neon is Google DeepMind's text model for complex software engineering, analytical knowledge work and science workflows. Work with text and media in a shared 1M-token context with four thinking levels. $1.65 / 1M input · $8.25 / 1M output.
- 1M-token context
- Long decode up to 1M
- Four thinking levels
- Function calling
- Multimodal input
Playground
Try native Gemini 4 Neon requests with your account balance. Select an available price tier and inspect the request before sending it.
gemini-4-neon$1.65 in · $8.25 out per 1M tokensMore Gemini parameters
Try one of these prompts
Messages are billed from your balance at the rates below, like an API call. Failed requests are free.
Gemini 4 Neon pricing
Gemini 4 Neon prices below come from current billing settings. Default uses the bare model ID, and additional tiers appear beside their full calling IDs. Thinking tokens count as output, even when the visible answer is short.
Example sizes are for illustration. Your bill follows the tokens each request reports. Current token rates. Cached input is billed once at its own rate; thinking counts as output. Check tool and processing-tier availability before using them.
| Category | Defaultgemini-4-neon | google llm vertextgemini-4-neon:googlellm-vertext |
|---|---|---|
| Price · 1M tokens | Price · 1M tokens | |
| Input | $1.65 | $2.40 |
| Output, including thinking | $8.25 | $12.00 |
| Cached input | $0.0825 | $0.12 |
What is Gemini 4 Neon?
Gemini 4 Neon is Google DeepMind's Gemini 4 model positioned between Gemini 4 Hydrogen and Gemini 4 Carbon for high-frequency complex tasks. It reads text, images, audio, video and PDFs, then returns text for software engineering, knowledge work, science and security workflows.
Gemini 4 Neon uses the same native generateContent request format as Hydrogen, Carbon and Argon, so migrating from another Gemini 4 model needs only a model ID update. Its default thinking level is high, with xhigh available when a hard problem needs deeper reasoning.
Google model specification, migration guide and v1beta discovery revision 20261119 checked November 20, 2026. Input, thinking and output share one 1,048,576-token window.
Two ways to use Gemini 4 Neon
Use Gemini 4 Neon from your application or coding agent.
Call it from your backend
Use the native Gemini request format for Gemini 4 Neon. The model ID belongs in the endpoint path.
- 1Create an API key
- 2Put gemini-4-neon in the endpoint path
- 3Send messages in contents and settings in generationConfig
Hand it to your coding agent
Copy the prompt, then review the request and estimated cost in your agent.
- 1Copy the prompt
- 2Paste it into your agent
- 3Approve the request
What Gemini 4 Neon is good at
Bring a multi-step engineering, analytical or scientific task with clear checks.
Repository-level debugging and refactoring
Give Gemini 4 Neon the failing tests, stack traces and related modules across a codebase. Ask it to trace root causes, plan clean multi-file edits and check edge cases before you merge.
Knowledge work and scientific synthesis
Gemini 4 Neon synthesizes dense PDFs, spreadsheets, charts and multi-step tool outputs inside its 1M-token context window into structured deliverables you can audit against source documents.
Long agentic tool sessions
Declare your native functions for Gemini 4 Neon and preserve returned thought signatures across turns. Your application executes each tool call and supplies the verified result on the next turn.
Long-decode deliverables up to 1M tokens
With continuationToken loops, Gemini 4 Neon can generate past the 65,536-token single-request cap, up to 1,048,576 tokens minus your input, without truncating long specifications.
Practical Gemini 4 Neon tasks
These are workflows to test with your own evaluation suite. No measured quality or speed advantage is claimed.
Audit concurrency and security bugs
Share lock acquisition code, retry queues and integration logs. Ask Gemini 4 Neon to locate race conditions, isolate root causes and draft a targeted patch you verify with tests.
Turn dense filings into structured JSON
Pass Gemini 4 Neon long financial reports, legal appendices or lab tables with a strict response schema. Verify extracted fields and source citations before writing to downstream storage.
Review technical recordings and specs
Attach a recorded session or architecture diagram with supporting docs. Ask Gemini 4 Neon to flag contradictions, missing steps and timestamps you can check directly.
How to use Gemini 4 Neon
- 01
Prepare your account
Sign in, add credit and create an API key for server-side requests. Check Gemini 4 Neon above to review the Default and explicit price tiers.
- 02
Assemble the conversation
Place user prompts, prior model turns and function results in contents. Attach images, audio, video or PDFs with accurate MIME types for Gemini 4 Neon.
- 03
Set thinking and output budget
Gemini 4 Neon defaults to high thinking. Pick low, medium, high or xhigh deliberately, and leave enough maxOutputTokens for both internal reasoning and final text.
- 04
Handle responses and continuations
Read Gemini 4 Neon candidates, finishReason and usageMetadata. Preserve complete model parts and thought signatures across turns, and pass continuationToken when needed.
The native Gemini 4 Neon API
Gemini 4 Neon uses the native generateContent and streamGenerateContent endpoints. Switching from Hydrogen, Carbon or Argon needs no request schema changes besides the model ID.
Use the stable model ID
Put gemini-4-neon directly in the endpoint URL. Gemini 4 Neon uses the same stable ID for JSON responses and SSE streaming, with optional :GROUP_NAME for explicit tiers.
Skip deprecated sampling fields
Gemini 4 Neon deprecates temperature, topP, topK and candidateCount, and does not support explicit caching. Implicit caching runs automatically from 4,096 prompt tokens.
Continue past 64K tokens
When a Gemini 4 Neon response ends with finishReason CONTINUATION, resend the exact original request with the returned continuationToken instead of appending partial text.
Gemini 4 Neon vs Hydrogen and Carbon
Gemini 4 Neon sits between Hydrogen and Carbon in the Gemini 4 lineup. Compare them using the same explicit thinking level and output budget on your own tasks.
| Setting | Gemini 4 Neon | Gemini 4 Hydrogen | Gemini 4 Carbon |
|---|---|---|---|
| Default thinking | high | high | medium |
| Thinking levels | low, medium, high, xhigh | low, medium, high, xhigh | low, medium, high |
| Shared context | 1,048,576 tokens | 1,048,576 tokens | 1,048,576 tokens |
| Non-streaming REST output per hop | 65,536 tokens | 65,536 tokens | 32,768 tokens |
| Request format | Native Gemini | Native Gemini | Native Gemini |
Test with identical prompts, schemas and explicit thinkingLevel settings. Model positioning in documentation does not replace running your own evaluation. Google Gemini 4 Neon guide
Gemini 4 Neon limits
One shared 1M window
Input, thinking and output share 1,048,576 tokens. Each Gemini 4 Neon request returns up to 65,536 tokens in both JSON and streaming modes; longer outputs require continuation requests.
Text output only
Gemini 4 Neon reads text, images, audio, video and PDFs and outputs text. It does not generate images or audio, and it does not support the Live API.
Account for thinking in your cap
Thinking tokens count against maxOutputTokens. Setting a tight limit can cut Gemini 4 Neon short before the final answer appears. Always inspect finishReason before retrying.
Verify built-in tools and tiers
Google lists Search, Maps, code execution, URL context, file search and computer use for Gemini 4 Neon. Account-level support for each tool and processing tier still requires live verification.
Use Gemini 4 Neon with SeedRouter
One balance across models
Pay for Gemini 4 Neon from the same SeedRouter balance as every other listed model. One-time top-up credit never expires.
Live rates beside the Playground
Check current Gemini 4 Neon rates directly on this page. The Playground calculates token estimates using the active price tier.
Failed requests are not charged
Failed Gemini 4 Neon requests cost nothing. Completed responses report prompt, cached input, candidate output and thinking tokens separately.
Explore other text models
Compare available text models by their documented capabilities and current rates.
Call GPT-6 Astra with the OpenAI SDK. OpenAI's most capable model, with a 1.05M-token context, live per-token pricing and a playground you can try right now.
View pricingCall GPT-6 Sol with the OpenAI SDK. 1.05M-token context, reasoning effort from none to max, live per-token pricing and a playground you can try right now.
View pricingCall GPT-6.1 Sol with the OpenAI SDK: near-Astra coding and agent work at a fifth of the price, 1.05M-token context, 95% off cached input. Try it now.
View pricingGPT-6.2 Sol is OpenAI’s everyday model for writing code and running agents at scale. It can interpret text and images, call tools and return structured results.
View pricingGPT-6.2 Astra is OpenAI’s flagship model for demanding reasoning and coding. It can build and revise software and work across documents in a single project.
View pricingCall GPT-6 Luna with the OpenAI SDK. The lowest-cost GPT-6 model, with a 1.05M-token context, live per-token pricing and a playground you can try right now.
View pricingGemini 4 Neon guides
Notes on Gemini 4 Neon parameters, thinking levels and model selection.
Gemini 4 Neon FAQ
What is Gemini 4 Neon?
Gemini 4 Neon is Google DeepMind's text model for complex software engineering, knowledge work, science and security workflows. It accepts text, images, audio, video and PDFs in a shared 1,048,576-token context and returns text.
Which Gemini 4 Neon model ID should I use?
Use the stable model ID gemini-4-neon in the native endpoint URL for the Default tier, or append :GROUP_NAME when selecting an explicit tier. Control reasoning depth with generationConfig.thinkingConfig.thinkingLevel.
How does Gemini 4 Neon differ from Hydrogen and Carbon?
Google positions Gemini 4 Neon between Hydrogen and Carbon for high-frequency complex tasks. Neon and Hydrogen both default to high thinking, support xhigh and return up to 65,536 tokens per non-streaming REST hop, while Carbon defaults to medium and caps at high.
What thinking levels does Gemini 4 Neon support?
Gemini 4 Neon supports low, medium, high (default) and xhigh thinking levels. Thinking cannot be disabled and minimal is rejected. Start with high for complex work, drop to low or medium for faster responses, and reserve xhigh for the hardest multi-step problems.
How long can a Gemini 4 Neon response be?
One Gemini 4 Neon hop returns up to 65,536 tokens in both JSON and streaming modes. With long decode and continuationToken requests, total output can reach 1,048,576 tokens minus your input. Each continuation request is billed separately and is not triggered automatically in the Playground.
How is Gemini 4 Neon billed?
Gemini 4 Neon bills input, cached input and output tokens, with thinking tokens counted as output. The pricing section above reads live SeedRouter rates directly from billing settings rather than hard-coding launch prices. Failed requests are not charged.
How do Gemini 4 Neon price tiers work?
Default uses the bare gemini-4-neon ID. Explicit tiers use the complete gemini-4-neon:GROUP_NAME ID shown next to their rates. The Playground lists only tiers available to your account, and an explicit selection stays within that tier.
Try Gemini 4 Neon
Bring a complex coding, analysis or document task you can verify. Try Gemini 4 Neon, inspect the native response and token usage, and keep the settings that fit your workload.
