Google Introduces Gemini 4 Neon (November 2026)

Abstract mid-weight glowing node between a lighter fast path and a heavier long-horizon block, illustrating Gemini 4 Neon

Written by

in

Takeaway: On November 18, 2026, Google introduced Gemini 4 Neon, a new Gemini 4 model aimed at frontier-level work that still needs to finish quickly — complex software engineering, knowledge work, and science workflows. The stable API model code is gemini-4-neon. Official paid Gemini API pricing is $3 per million input tokens and $15 per million output tokens, including thinking tokens, with implicit context caching at $0.15 per million tokens. The model is not available on the Gemini API free tier.

The details below come from that announcement by Koray Kavukcuoglu, SVP at Google DeepMind and Google’s chief AI architect, the Gemini API models page, the Developer API pricing page, and the Gemini 4 Neon model card published November 18, 2026. This is a new model in the Gemini 4 family, not a feature change like this week’s Gemini Tasks rollout for free users in six countries or the updated Gemini chat history and training controls.

Where Neon sits next to Carbon, Argon, and Hydrogen

Google’s launch post and model card use the same positioning. Carbon is for the hardest, long-horizon projects. Hydrogen is the fast everyday execution model — the one we covered in our Gemini 4 Hydrogen launch article. Neon, Google says, “balances depth and efficiency” and is built for complex coding, analytical reasoning, and production-grade enterprise workflows. The model card also states that Gemini 4 Neon is adapted from Gemini 4 Carbon.

That does not make Neon a replacement for Carbon, and Google’s announcement does not describe it as a replacement for Gemini 4 Argon either. Argon remains listed on the official models and pricing pages as the efficient Gemini 4 frontier reasoning model. If you need the deepest long-horizon work, Carbon is still the model Google points to, including in our earlier coverage of Carbon’s November 10 general availability.

Price, context window, and what the API actually bills

Google’s announcement and the current pricing page agree on the paid list price: $3 per million input tokens and $15 per million output tokens. Cached input is $0.15 per million tokens, which matches the 95% cached-input discount in the launch post. The pricing page says implicit caching is the only caching option; explicit caching is not supported. Output pricing includes thinking tokens.

Two billing details are easy to miss. First, Neon is not available on the free tier. Second, the pricing page says every token the model processes is counted and billed, including thinking tokens and internal tool system instructions and function declarations when using grounding. Paid-tier content is listed as not used to improve Google’s products.

The token price is not the same thing as Google’s cost-per-task claims. Neon’s list price is higher than Carbon’s current introductory rate of $2 / $10, which the pricing page says lasts through November 30, 2026, before moving to $4 / $20 on December 1, 2026. Google’s launch post instead talks about cost per task: at Neon’s default high thinking level, it says the model has a 32% lower cost per task than Carbon, and that at the xhigh thinking level Neon still costs 36% or more less per task than competing models at their highest setting. Those figures are Google’s, not independent invoice math.

The models page and model card describe a 1 million-token context window. Default synchronous output is up to 64K tokens, with up to 1 million tokens via background long decode. Inputs are text, images, audio, video, and PDF. The model card lists a knowledge cutoff of March 2026, while noting that knowledge in some domains may be less up to date.

Who can use it today

Google says Gemini 4 Neon began rolling out on November 18, 2026, on these paths:

  • Developers: the Gemini API paid tier in Google AI Studio and Android Studio. Paid Google AI Pro or Ultra subscribers can try Neon in Google Antigravity, and all Pro and Ultra subscribers can try it in AI Studio.
  • Enterprises: Gemini for business, via the API or the model drop-down, and Google Antigravity.
  • Consumers: Google AI Pro or Ultra subscribers can select Gemini 4 Neon from the model drop-down in the Gemini app.
  • Cybersecurity defenders: eligible Fairwind Program members “will begin receiving access over the next few days.” Google tells people to request Fairwind access through their Cloud account team. This is the same limited program we described in our Carbon benchmarks and Fairwind coverage.

Google has not said that free Gemini app users can pick Neon from the model drop-down. The launch post does say that Tasks in the Gemini app can now run on Neon. That is a runtime change for Tasks, not a claim that the consumer model picker is open to every account.

Thinking levels and the official benchmark picture

Neon has four thinking levels. Google says high is the default, and that xhigh adds deeper, multi-step reasoning for the hardest engineering, analytical, and scientific workloads. The headline Artificial Analysis Intelligence Index score in the launch post — 61, ranked third overall — is at the xhigh thinking level. FrontierCode 2.0 is reported at 46.2% at the default high setting.

Other official scores from the launch post and the November 2026 model card include: 70.9% and a number-two rank on the Vals Index; 65.2% on Terminal-Bench 4.1; 58.6 and a number-one rank on the Artificial Analysis Coding Agent Index; 75.8% on Agents’ Last Exam v1.1; 56.3% and a number-one rank on Zapier’s AutomationBench; $16,843 and a number-one rank on Andon Labs’ Vending-Bench 2; 73.5% on Chartography; and, on science and math, 70.5% on Terminal-Bench Science, 84.0% on RiemannBench, and 71.3% on HLE-Diamond. Google also says Neon ranks first on the Legal Research Bench and the Tax Agent Bench, and number two on steerability in the Agent Arena Index, with a global Agent Arena Index score of 1336.

On SWE-sweep, the launch post and model card both report 12.8% on the mini-SWE harness and 39.0% in Google Antigravity. That gap is Google’s own illustration of what the Antigravity harness adds; it is not a promise that every coding setup will match the higher figure.

Cyber defense stays behind Fairwind

Google describes a separate “Gemini 4 Neon for cyber” configuration for trusted defenders, with official scores on AutoBaxBench, WebExploitBench v1.2, SEC-bench Pro (90.5%), ExploitGym, and Collinear’s CWE-bench v2 (70.7%). Access is not automatic with the public API or the Gemini app. Google says this configuration is rolling out to trusted defenders in the coming days as part of the Fairwind Program.

The model card’s full Safety & Security Report is not published yet. Google says it will be available in the coming days. Until that report is out, treat the cyber and safety figures above as the company’s launch and model-card statements, not as a completed public safety report.

Key facts

ItemWhat Google says
AnnouncementNovember 18, 2026 blog post, “Introducing Gemini 4 Neon,” by Koray Kavukcuoglu. Model card published the same day.
Model IDgemini-4-neon
PositioningBalances depth and efficiency between Hydrogen (fast everyday execution) and Carbon (hardest long-horizon work). Adapted from Gemini 4 Carbon.
Paid API price$3 per million input tokens, $15 per million output tokens (including thinking tokens). Implicit context caching $0.15 per million tokens. Not available on the free tier.
Context and output1 million-token context window. Up to 64K default synchronous output tokens, or up to 1 million via background long decode. Inputs: text, image, audio, video, and PDF.
Thinking levelsFour levels; high is the default. The AA Intelligence Index score of 61 is at xhigh.
Consumer and developer accessPaid Gemini API in Google AI Studio and Android Studio; Gemini for business; Google Antigravity for paid Pro and Ultra subscribers; Gemini app model drop-down for Pro and Ultra subscribers.
Cyber configurationNeon for cyber rolls out to eligible Fairwind Program members over the next few days, not automatically through the public API.
Knowledge cutoffMarch 2026, according to the model card, with the caveat that some domains may be less current.

FAQ

What is Gemini 4 Neon?

It is a Gemini 4 model Google introduced on November 18, 2026, for complex software engineering, knowledge work, and science workflows. Google positions it between Hydrogen, the fast everyday model, and Carbon, the model for the hardest long-horizon jobs.

What is the API model name and price?

The stable model code is gemini-4-neon. Official paid pricing is $3 per million input tokens and $15 per million output tokens, including thinking tokens, with implicit context caching at $0.15 per million tokens. It is not on the Gemini API free tier.

Can free Gemini app users select Neon?

Google’s launch post says Google AI Pro or Ultra subscribers can select Neon from the model drop-down in the Gemini app. It does not say free users can pick the model themselves. The post does say Tasks in Gemini can now run on Neon.

Is Neon cheaper than Gemini 4 Carbon?

Not on the current list price. Neon is $3 / $15 per million tokens. Carbon is still at its introductory $2 / $10 rate through November 30, 2026, then $4 / $20 from December 1. Google separately says Neon has a lower cost per task than Carbon at the default high thinking level. That is a task-cost claim, not the per-token list price.

Does Neon replace Gemini 4 Argon or Carbon?

Google’s announcement does not say that. Carbon remains the model for the hardest long-horizon projects, and Argon remains listed on official models and pricing pages. Neon is described as the balance of depth and efficiency, adapted from Carbon.

How large is the context window?

Google lists a 1 million-token context window, with up to 64K default synchronous output tokens and up to 1 million tokens through background long decode. Supported inputs are text, images, audio, video, and PDF.

What is Gemini 4 Neon for cyber?

It is a cybersecurity configuration Google is rolling out to trusted defenders in the Fairwind Program over the next few days. It is not the same as getting gemini-4-neon in the public API or the Gemini app. Access requests go through a Cloud account team.

Where should I confirm the latest details?

Start with Google’s November 18 announcement, the Gemini API models and pricing pages, and the official model card. The full Safety & Security Report is still listed as coming in the coming days.

Comments

Leave a Reply