Skip to content

Gemini 4 Neon explained: price, benchmarks and access.

Google’s Gemini 4 Neon sits between Hydrogen and Carbon at $3/$15 per million tokens with a 1M context. Benchmarks, thinking levels and how to try it.

By The Model Atlas · 4 sources

In this post9 sections
  1. 01What is Gemini 4 Neon?
  2. 02Gemini 4 Neon benchmarks
  3. 03What’s new for coding and agents
  4. 04Gemini 4 Neon for cyber defense
  5. 05Gemini 4 Neon pricing
  6. 06How to try Gemini 4 Neon
  7. 07Who is Gemini 4 Neon for?
  8. 08What people are building with Gemini 4 Neon
  9. 09Gemini 4 Neon FAQ

Gemini 4 Neon is Google’s newest Gemini 4 model, announced on 18 November 2026. Google describes Gemini 4 Neon as bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work and science workflows. It is rolling out in the Gemini API, Google AI Studio, Google Antigravity, Gemini for business and the Gemini app. Here is what Gemini 4 Neon is, what Google says it can do, what it costs and how to try it.

What is Gemini 4 Neon?

Neon is the middle model of the Gemini 4 family. Google’s developer docs say it sits between Gemini 4 Hydrogen and Gemini 4 Carbon for high-frequency complex tasks. In Google’s words, Carbon tackles your hardest, long-horizon projects, Hydrogen handles fast everyday execution, and Neon balances depth and efficiency.

The core specs, from Google’s Gemini API and Google Cloud documentation:

  • Model ID gemini-4-neon, on the generateContent and Interactions APIs (Google Cloud also lists Count Tokens and Chat Completions support).
  • A unified 1,048,576-token context window shared by input, thinking and output tokens, with long decode up to 1M tokens.
  • Synchronous outputs up to 65,536 tokens per request; longer outputs run through background execution or client-side continuation.
  • Text, image, audio and video input, with text output.
  • Four thinking levels: low, medium, high (the default) and xhigh.
  • Paid tier only: Google says Neon is not available on the Gemini API free tier.

Gemini 4 Neon benchmarks

Google’s announcement makes these claims about Neon (all figures are Google’s, from its 18 November post):

  • Artificial Analysis Intelligence Index: #3 overall with a score of 61 at the xhigh thinking level, which Google says puts it just behind larger, slower models.
  • Artificial Analysis Coding Agent Index: #1 overall at 58.6.
  • Terminal-Bench 4.1: 65.2%. FrontierCode 2.0: 46.2% at the default high setting.
  • SWE-sweep: 12.8% on the mini-SWE harness, rising to 39.0% inside Google Antigravity, which Google calls state of the art.
  • Vals Index: #2 at 70.9%, and #1 on both Legal Research Bench and Tax Agent Bench.
  • Agent Arena Index from Arena: #2 globally with a score of 1336 across 4,200+ real-world agent sessions, and #2 on steerability.
  • Zapier AutomationBench: #1 at 56.3%. Andon Labs Vending-Bench 2: #1 at $16,843.
  • Science and math: 70.5% on Terminal-Bench Science, 84.0% on RiemannBench and 71.3% on HLE-Diamond.
  • Agents’ Last Exam v1.1: 75.8%, which Google describes as on par with leading models. Chartography: 73.5%.

Google also says Neon recorded its lowest fabrication rate yet in internal Google Antigravity testing, and that at the default high thinking level it has a 32% lower cost per task than Carbon.

What’s new for coding and agents

Google says Neon is engineered for long-running agent workflows. It can run on its own or act as a fast Orchestrator directing parallel Hydrogen subagents in Google Antigravity. According to Google, it cuts down on redundant file reads, shares progress as it works, handles context compaction reliably in long sessions, and does multi-file debugging and refactoring with targeted edits.

The developer docs add that the Antigravity agent in Gemini Managed Agents supports gemini-4-neon, either standalone or paired with Hydrogen in multi-model workflows.

Gemini 4 Neon for cyber defense

Google says Neon writes more secure code than Carbon across all difficulty tiers of AutoBaxBench, with a state-of-the-art 76% on the hard tier. It reports 51.26% on WebExploitBench v1.2, 90.5% on SEC-bench Pro and 70.7% on Collinear’s CWE-bench v2. A Gemini 4 Neon for cyber version is rolling out to trusted defenders in the coming days through Google’s Fairwind Program; Google says to contact your Cloud account team to request access.

Gemini 4 Neon pricing

Google’s Gemini API pricing for Neon is flat across the full 1M context window:

  • Input: $3.00 per million tokens.
  • Cached input: $0.15 per million tokens. Neon supports implicit caching only, which Google says discounts cached input by 95%; explicit caching is not supported.
  • Output, including thinking tokens: $15.00 per million tokens.
  • Grounding with Google Search: 5,000 free requests a month shared across Gemini 3 and newer models, then $14 per 1,000 requests.

Google notes that every token the model processes is billed, including thinking tokens and internal tool instructions when grounding is used.

How to try Gemini 4 Neon

  • Developers: use the Gemini API paid tier in Google AI Studio and Android Studio. All Google AI Pro and Ultra subscribers can try it in AI Studio.
  • Google Antigravity: available to paid Google AI Pro and Ultra subscribers, and to enterprises.
  • Enterprises: Gemini for business, through the API or the model drop-down.
  • Consumers: Google AI Pro and Ultra subscribers can pick Gemini 4 Neon from the model drop-down in the Gemini app.
  • Cyber defenders: eligible Fairwind Program members get access over the next few days.

If you already use another Gemini 4 model, Google’s migration guide says switching needs no request schema changes: set the model to gemini-4-neon and check the thinking level, since Neon defaults to high and supports xhigh while Carbon and Argon default to medium and cap at high. Google also says Neon ignores temperature, top_p and top_k and rejects the minimal thinking level.

Who is Gemini 4 Neon for?

Based on Google’s positioning, Neon targets teams that want near-Carbon quality on complex but frequent work: repository-level coding, agent workflows, spreadsheet and document analysis, legal, finance and science tasks. Google recommends keeping Hydrogen as the default for everyday coding and agentic work and using Carbon for the hardest architecture, debugging, science and security problems.

What people are building with Gemini 4 Neon

Google’s own launch demos, all built in Google Antigravity, include a first-person WebGL2 freeride ski simulator on real Mont Blanc and Chamonix terrain with 10,200 lines of custom shaders, an interactive 3D recreation of Alan Turing’s 1936 study at King’s College, Cambridge, and a 100,000-cell Google Sheet that uses over 500,000 formulas to relight a scene from a single image.

On The Model Atlas, an early community build pits Neon against Carbon on a one-file Three.js Earth zoom: see Earth-to-Statue of Liberty zoom: Gemini 4 Neon vs Carbon. You can follow new Neon builds on the Gemini 4 Neon model page and compare with builds made with Gemini 4 Carbon, Gemini 4 Hydrogen and Gemini 4 Argon.

Gemini 4 Neon FAQ

When was Gemini 4 Neon released?

Google announced Gemini 4 Neon on 18 November 2026 and began rolling it out the same day.

How much does Gemini 4 Neon cost?

In the Gemini API, Google charges $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens, including thinking tokens. It is paid tier only.

Is Gemini 4 Neon free?

Not in the API: Google says Neon is not available on the Gemini API free tier. In the Gemini app and Google Antigravity it is offered to Google AI Pro and Ultra subscribers.

What is Gemini 4 Neon’s context window?

Google lists a unified 1,048,576-token window shared by input, thinking and output, with up to 65,536 output tokens per synchronous request and long decode up to 1M tokens.

How is Gemini 4 Neon different from Gemini 4 Carbon and Hydrogen?

Google places Neon between them: Carbon for the hardest long-horizon problems, Hydrogen for fast everyday execution, and Neon for complex, high-frequency work. Google says Neon has a 32% lower cost per task than Carbon at the default high thinking level.

What thinking levels does Gemini 4 Neon support?

Four: low, medium, high (the default) and xhigh. Google recommends starting with high and reserving xhigh for hard multi-step problems.

Sources · 4

  1. Introducing Gemini 4 Neon (Google, 18 November 2026) (opens in a new tab)
  2. What’s new in Gemini 4 Neon (Gemini API docs) (opens in a new tab)
  3. Gemini Developer API pricing (Google AI Studio) (opens in a new tab)
  4. Gemini 4 Neon model page (Google Cloud documentation) (opens in a new tab)
All 7 builds →