Gemini 4 Neon is Google's new middle model in the Gemini 4 family, released on November 18, 2026. It is adapted from Gemini 4 Carbon, costs $3 per million input tokens and $15 per million output tokens on the paid Gemini API, and is the model Google AI Pro subscribers ($19.99 a month) can now pick in the Gemini app as "Deeper analysis". It is not on the free API tier or the free and Plus app plans.

In short, Neon gives Pro subscribers a Carbon based model without Ultra, and it scores 61 on the Artificial Analysis Intelligence Index, one point above Carbon's default setting. The catch: it writes many thinking tokens, so real task costs run above the token price.

Facts below are checked against Google's announcement, the Neon model card, the Gemini API docs and pricing page, Google's plan and help pages, and Artificial Analysis on November 19, 2026. We have not run Neon ourselves.

What is Gemini 4 Neon?

Koray Kavukcuoglu, senior vice president at Google DeepMind and Google's chief AI architect, introduced Neon in a Google blog post on November 18. Google pitches it for complex coding, analytical reasoning and enterprise work that has to finish quickly. In its own words, Carbon handles the hardest long projects, Hydrogen handles fast everyday work, and Neon "balances depth and efficiency" between them.

The Gemini 4 Neon model card from Google DeepMind gives the basics:

  • Built from: Gemini 4 Carbon.
  • Inputs: text, images, audio, video and PDF. Output is text only, with no image or audio generation.
  • Context: one window of 1,048,576 tokens shared by input, thinking and output.
  • Output length: 65,536 tokens per normal response, and up to the full window with background long decode on the API.
  • Knowledge cutoff: March 2026.
  • Thinking levels: four (low, medium, high and extra high), with high as the default, per 9to5Google and Artificial Analysis.

The API model ID is gemini-4-neon, listed as stable on the Gemini API model page, with function calling, structured outputs, code execution, computer use and Google Search grounding. It does not support the Live API.

Who can use Gemini 4 Neon?

Neon reached paying users on day one, with no cyber defenders first period.

WhereWho gets NeonNotes
Gemini appGoogle AI Pro and UltraPick "Deeper analysis" in the model menu; also runs Tasks
Gemini app, Free and PlusNoAuto and Hydrogen only
Gemini APIPaid tier onlyNot available on the free tier
Google AI StudioPaid API users; Pro and Ultra subscribers can try itModel ID gemini-4-neon
Google AntigravityPaid Google AI Pro and UltraCan direct parallel Hydrogen subagents
Gemini for businessYesAPI and model menu
Neon for cyberFairwind Program membersRolling out "in the coming days"

The app lineup is the bigger change for most people. Google's Gemini Apps help page on models now lists three models: Hydrogen for everyday help, Neon for deeper analysis and Carbon for the hardest problems. Free users get Auto in chats, Plus adds Hydrogen by name, Pro adds Neon, and only Ultra gets Carbon. 9to5Google reports that Neon replaces Argon and the older 3.1 Pro in the app picker. Google's subscriptions page now lists "Access to Gemini 4 Neon" as the headline benefit of the $19.99 Pro plan, while Ultra (from $99.99 a month) keeps Carbon.

Picking Neon by hand uses more of your plan's allowance, and when it runs out, chats continue on a lighter model. For a full plan comparison with ChatGPT Plus and Claude Pro, see ChatGPT Plus vs Claude Pro vs Google AI Pro.

Gemini 4 Neon price

Google's Gemini API pricing page lists one flat price across the whole context window, with no higher rate for long prompts:

TierInput, per 1M tokensOutput, per 1M tokensCached input
Standard$3.00$15.00$0.15
Batch$1.50$7.50$0.075
Flex$1.50$7.50$0.075
Priority$5.40$27.00$0.27
Free tierNot availableNot availableNot available

Output includes thinking tokens. Caching is implicit only, at a 95% discount. Search grounding is free for 5,000 requests a month, then $14 per 1,000.

How that compares inside the family:

  • Hydrogen costs $1.50 and $7.50, half Neon's rate, and it is on the free tier.
  • Carbon and Argon cost $2 and $10 through November 30, 2026, so Neon is actually dearer per token than Carbon for the rest of this month.
  • From December 1, Carbon and Argon rise to $4 and $20, and Neon becomes 25% cheaper than both.

Per token price is only half the bill. Artificial Analysis measured Neon at extra high writing 283 million output tokens across its index, against a median of 65.1 million for models in its price range, and rated it "very verbose". That put its cost at $6.44 per index task and its time at 21.5 minutes per task, slower than the median of 12.1 minutes. At the default high setting Neon used 123 million output tokens, less than half the extra high figure, and Google says high costs 32% less per task than Carbon. Our LLM subscription vs API cost comparison shows what these rates mean in practice.

Gemini 4 Neon vs Carbon and Hydrogen

ItemGemini 4 HydrogenGemini 4 NeonGemini 4 Carbon
Released to the publicNovember 12, 2026November 18, 2026November 10, 2026
Google's role for itFast everyday workDepth with speedHardest long tasks
Gemini app planPlus and upPro and upUltra only
API price per 1M tokens$1.50 / $7.50$3 / $15$2 / $10, then $4 / $20 from Dec 1
Free API tierYesNoNo
Artificial Analysis index51 high, 54 extra high60 high, 61 extra high60 high
Context windowAbout 1MAbout 1MAbout 1M

The scores come from Artificial Analysis model pages for Neon and Carbon, plus our Gemini 4 Hydrogen explainer. Neon's 61 at extra high is the best Google score on that index so far, and Google calls it third overall at launch, behind larger models. On the index home page on November 19, Claude Fable 5.2 (max) led at 65 and GPT-6.2 Astra (max) was second at 62.

So which Gemini should you use? Neon at its default high setting matches Carbon's index score at a lower cost per task. Google still keeps Carbon for the hardest, longest projects, and our Gemini 4 Carbon explainer covers what Carbon is built for. Hydrogen stays the cheap, fast choice and the only one on the free API tier. If you came here looking for Argon, our Gemini 4 Argon release date page covers it; Argon is still on the API but no longer in the app's model list.

Gemini 4 Neon vs Claude Opus 5.5 and GPT-6.2 Sol

Google chose Claude Opus 5.5 and GPT-6.2 Sol as the comparison models on Neon's model card. Here is how list prices line up, from the Claude pricing page and OpenAI API pricing:

ModelInput, per 1MOutput, per 1M
Gemini 4 Neon$3$15
Claude Opus 5.5$4$20
Claude Sonnet 5.5$2$10
GPT-6.2 Sol$2$10
GPT-6.2 Astra$8$40
Claude Fable 5.2$8$40

Neon is 25% cheaper per token than Opus 5.5 and 50% dearer than Sonnet 5.5 and GPT-6.2 Sol. VentureBeat made the same point and added that Neon needs about a third fewer tokens than Sol or Sonnet to cost the same per job.

On Google's own model card table, Neon leads Opus 5.5 and GPT-6.2 Sol on most rows: 71.3% against 67.9% and 64.0% on HLE-Diamond, and 95.2% against 66.8% and 80.1% on the GraphWalks long context test at 256K to 1M tokens. It trails Opus 5.5 slightly on Terminal-Bench 4.1 (65.2% vs 65.8%) and FrontierCode 2.0 (46.2% vs 47.3%), and trails GPT-6.2 Sol on Chartography (73.5% vs 75.1%). These are Google's numbers. How the leaderboards work is covered on our sister site's Artificial Analysis Intelligence Index guide. For the rivals themselves, see GPT-6 Sol vs Astra vs Luna and Claude Fable 5.2 explained, or use Claude and ChatGPT directly to compare on your own tasks.

What Google says about Neon's safety

First, Neon did not get a full new round of dangerous capability testing. Under its Frontier Safety Framework, Google ran a Material Capability Change check against Gemini 4 Argon, the last model that had full critical capability evaluations. Neon showed a capability increase, but below Google's threshold, so Google carried Argon's findings forward. Those findings say a severe biological uplift level and a severe cyber uplift level "cannot be ruled out", and that a loss of control warning level for stealth and situational awareness "cannot be ruled out". Google says mitigations are deployed for each. The full Neon Safety and Security Report is listed as "coming soon".

Second, Google's announcement reports these behavior results:

  • 0.0% cheating on Terminal-Bench 4.0, matching Carbon.
  • Shortcuts on 2.45% of tasks in its coding cheating test, the lowest of any Gemini 4 model.
  • A 0.09% misaligned action rate on its agentic safety suite and a 99.2% pass rate on its agent honesty test.
  • Against Hydrogen, unjustified refusals fell 3.00 points and image safety rose 1.90 points, while multilingual safety slipped 0.40 points.

All of these are Google's internal, automated tests. The card says plainly that the safety results are not human evaluation or red teaming. For how Google's framework compares with OpenAI's and Anthropic's rules, see Preparedness Framework vs Responsible Scaling Policy.

Pros and cons of Gemini 4 Neon

Pros

  • Carbon based model on the $19.99 Pro plan, no Ultra needed.
  • Ties Carbon's index score at its default setting and beats it at extra high.
  • One flat API price across the full 1M token window, with half price batch and flex.
  • 25% cheaper per token than Carbon and Opus 5.5 once Carbon's intro price ends on December 1.
  • Takes text, images, audio, video and PDFs in one prompt.

Cons

  • Not on the free API tier or the free and Plus app plans.
  • Very verbose at extra high, so real costs per task are higher than the token price suggests.
  • Slower per task than most models in its price range, per Artificial Analysis.
  • No full new dangerous capability evaluation; the Safety and Security Report is not out yet.
  • Text output only.

Should you switch to Gemini 4 Neon?

  • You pay for Google AI Pro: yes, try it for long documents, spreadsheets and research. It is the strongest model your plan offers.
  • You are on Free or Plus: stay on Hydrogen unless it falls short on hard work.
  • You build on the API: test Neon at high before extra high, and compare cost per finished task, not cost per token, against Carbon, Sonnet 5.5 and GPT-6.2 Sol. If you are new to the API, our Google AI Studio vs Gemini guide explains the two surfaces, and Google AI Studio is where you get a key.
  • You need the top score: Claude Fable 5.2 and GPT-6.2 Astra still lead, at over twice Neon's token price.

FAQ

Is Gemini 4 Neon free?

No. Neon is not on the free Gemini API tier, and in the Gemini app it needs Google AI Pro ($19.99 a month) or Ultra. Free users get Auto, and Google AI Plus adds Hydrogen. On the paid API it costs $3 per million input tokens and $15 per million output tokens.

Is Gemini 4 Neon better than Gemini 4 Carbon?

On the Artificial Analysis Intelligence Index, Neon scores 61 at extra high and 60 at high, against 60 for Carbon at high, and Google says Neon costs 32% less per task at high. Google still positions Carbon for the hardest, longest projects, and Carbon remains an Ultra only model in the app.

Does Gemini 4 Neon replace Gemini 4 Argon?

In the Gemini app, yes. Google's help page now lists only Hydrogen, Neon and Carbon, and 9to5Google reports that Neon replaced Argon and 3.1 Pro in the model picker. Argon is still listed on the Gemini API pricing page.

What is the Gemini 4 Neon context window?

1,048,576 tokens, shared by input, thinking and output. A normal response can be up to 65,536 tokens, and the API's background long decode can run output up to the full window.

When was Gemini 4 Neon released?

November 18, 2026, on the paid Gemini API, Google AI Studio, Gemini for business, Google Antigravity and the Gemini app for Pro and Ultra subscribers. A cyber version reaches Fairwind Program members over the following days.