Gemini 4 Neon: Power-User Guide to Pricing, API and Thinking Levels

Last updated: November 19, 2026

Gemini 4 Neon (model ID gemini-4-neon) is Google’s new mid-tier Gemini 4 model, released November 18, 2026. It targets complex coding, knowledge work and science tasks, costs $3 per million input tokens and $15 per million output tokens on the paid API tier, and offers a 1 million-token context window with four thinking levels.

Contents

What Gemini 4 Neon is

In its launch announcement, Google describes Neon as a model that balances depth and efficiency. Carbon handles the hardest long-horizon projects, Hydrogen handles fast everyday work, and Neon covers complex tasks that run often enough that speed and cost matter.

The What’s new in Gemini 4 Neon page says the model completes repository-level debugging and multi-file refactoring in fewer steps, reduces redundant file reads in long sessions, and handles context compaction reliably.

According to the model page, Neon accepts text, images, video, audio and PDF files and returns text only. It supports code execution, computer use, function calling, file search, structured outputs, URL context, and grounding with Google Search and Google Maps. It does not support image generation, audio generation or the Live API.

Where you can use it

Google is rolling Neon out across several products at once:

  • Gemini API: paid tier only, through both generateContent and the Interactions API.
  • Google AI Studio: for Google AI Pro and Ultra subscribers.
  • Gemini app: Pro and Ultra subscribers can pick Gemini 4 Neon from the model menu. Google also says Tasks in Gemini can now run on Neon.
  • Google Antigravity: for paid Google AI Pro and Ultra subscribers, where Neon can work alone or act as an orchestrator directing parallel Hydrogen subagents.
  • Gemini for business: through the API or the model menu.

A separate cybersecurity version, Gemini 4 Neon for cyber, is rolling out to eligible members of Google’s Fairwind Program over the coming days. Access requests go through a Google Cloud account team.

Pricing and how it compares

The Gemini API pricing page lists flat pricing for Neon across the full context window. Output pricing includes thinking tokens, and Google also bills internal tool instructions and function declarations when grounding is used.

Model Input (per 1M tokens) Output (per 1M tokens) Cached input Free API tier
Gemini 4 Neon $3.00 $15.00 $0.15 No
Gemini 4 Carbon $2.00 through Nov. 30, 2026; $4.00 from Dec. 1 $10.00 through Nov. 30; $20.00 from Dec. 1 $0.10, then $0.20 No
Gemini 4 Argon $2.00 through Nov. 30, 2026; $4.00 from Dec. 1 $10.00 through Nov. 30; $20.00 from Dec. 1 $0.10, then $0.20 No
Gemini 4 Hydrogen $1.50 $7.50 $0.075 Yes

Until November 30, Neon costs more per token than Carbon and Argon because those models are still on introductory pricing. From December 1, Neon will be cheaper per token than both. Google Search grounding includes 5,000 free requests per month shared across Gemini 3 and newer models, then costs $14 per 1,000 requests.

How to call Gemini 4 Neon in the API

These steps follow Google’s Neon quickstart and migration guide:

  1. Use a Gemini API project on the paid tier. Neon requests will not work on the free tier.
  2. Install or update the Google GenAI SDK for Python (google-genai).
  3. Create a client and send a request with model="gemini-4-neon", for example through client.interactions.create().
  4. Leave the thinking level at the default high at first, or set "thinking_level" in generation_config.
  5. Remove temperature, top_p and top_k from your requests, and declare tools natively rather than asking for raw JSON tool calls.
  6. Set HTTP client timeouts to at least 15 minutes, or use background execution, because hard prompts can take several minutes.

Thinking levels and long outputs

The thinking level is the main control over quality, latency and token use. Neon supports four levels:

Level Google’s guidance
low Fewest tokens and lowest latency, for fast chat and high-throughput processing
medium Interactive coding, live UI prototyping and latency-sensitive tool calls
high (default) Recommended for most tasks, including complex coding and agent work
xhigh Hard multi-step engineering, analytical and science problems where high falls short

Google suggests running evaluations at all four levels with a high max_output_tokens value and then choosing the lowest level that meets your quality bar. The minimal level used by some earlier models returns an error.

Input, thinking and output share one 1,048,576-token window. A single synchronous response is capped at 65,536 tokens. For longer output, set max_output_tokens above 65,536 and set background to true, so the server continues in 64K-token steps. If you enable background mode without raising max_output_tokens, the response still stops at 65,536 tokens.

Migrating from other Gemini models

Moving from Gemini 4 Carbon, Gemini 4 Argon or Gemini 4 Hydrogen needs no request schema changes. Update the model ID and check the thinking level: Neon defaults to high and supports xhigh, while Carbon and Argon default to medium and stop at high.

From Gemini 3.8 Flash or older models, there is more to change. Remove explicit context caching fields, because Neon supports only implicit caching, which applies by default to prompts of at least 4,096 tokens. Remove sampling parameters, raise any hardcoded low output caps such as 8,192, and account for the shared context window.

If you use Neon inside Antigravity, our guides to the Antigravity Orchestrator and Gemini Tasks explain the surrounding features. To compare subscription plans, see Google AI Pro vs. Ultra.

Benchmarks Google reports

Google’s announcement lists several results. These are vendor-reported or third-party figures cited by Google, not independent tests by FreeAnswers:

  • Score of 61 on the Artificial Analysis Intelligence Index at xhigh, ranked third overall.
  • 58.6 on the Artificial Analysis Coding Agent Index, ranked first.
  • 65.2% on Terminal-Bench 4.1 and 70.9% on the Vals Index, ranked second.
  • At the default high level, a 32% lower cost per task than Carbon, according to Google.

Limitations and tips

  • No free API tier. Use Hydrogen if you need free-tier access.
  • Text output only. For images, audio or live voice sessions, you need a different model.
  • Thinking tokens cost money. They are billed at the $15 output rate, so lower the thinking level before cutting max_output_tokens.
  • New file inputs. The same Gemini API release notes entry adds direct support for Word, PowerPoint and Excel files, with page, tab, cell and time-range slicing so you process only what you need.

Frequently asked questions

What is Gemini 4 Neon?

Gemini 4 Neon is a Google model released on November 18, 2026, for complex coding, knowledge work and science tasks. It sits between Gemini 4 Hydrogen and Gemini 4 Carbon and aims to deliver frontier-level quality with faster task completion.

How much does Gemini 4 Neon cost?

On the paid tier of the Gemini API, Gemini 4 Neon costs $3.00 per 1 million input tokens and $15.00 per 1 million output tokens, including thinking tokens. Cached input from implicit caching costs $0.15 per 1 million tokens. Pricing is flat across the full context window.

Is Gemini 4 Neon free?

Not in the API. Google says Gemini 4 Neon is available on the paid tier only and not on the free tier. In the Gemini app, it appears in the model menu for Google AI Pro and Ultra subscribers.

What is the Gemini 4 Neon model ID and context window?

The model ID is gemini-4-neon. It has a unified 1,048,576-token context window shared by input, thinking and output tokens, with a default synchronous output limit of 65,536 tokens and long decode up to the full window.

Should I use Neon, Carbon or Hydrogen?

Google recommends Neon for complex, high-frequency coding, knowledge work and science tasks, Carbon for the hardest architecture, debugging, science and security problems, and Hydrogen as the default for everyday coding and agent work.

Sources

Image: Gemini 4 Neon announcement header, Google, used for editorial purposes.

Related answers

Leave a Comment