LATEST
AI Tools

Gemini 4 Neon Is Live: $3/$15, Carbon-Level Coding, Flash Speed

Updated November 19, 2026
13 min read
Editor's Note: This article was human-reviewed for accuracy
Vibrant neon light patterns swirling on a dark background

Quick Answer & Key Takeaways

  • • Google launched Gemini 4 Neon on November 18, 2026, priced at $3 per million input tokens and $15 per million output tokens, with a 1-million-token context window.
  • • Google's announcement puts Neon ahead of comparable models on coding and agent benchmarks, and claims 32% lower cost per task than Carbon at its default thinking level.
  • • From December 1, Carbon and Argon rise to $4/$20 per million tokens, which leaves Neon 25% cheaper per token than either of them.

Google has a fourth Gemini 4 model, and it is the one most developers will actually use. Gemini 4 Neon launched on November 18, 2026, priced at $3 per million input tokens and $15 per million output tokens, with a 1-million-token context window. Google positions it between Hydrogen, the fast everyday model, and Carbon, the flagship for the hardest work. The pitch is straightforward: Carbon-level results on coding and agent tasks, at a lower cost per task and with faster turnaround.

Neon matters for a second reason that has nothing to do with benchmarks. On December 1, the introductory pricing on Carbon and Argon ends, and both jump to $4 per million input tokens and $20 per million output tokens. Neon stays at $3 and $15. From that date, Neon is 25% cheaper per token than either of its siblings, while Google claims it matches or beats Carbon on the work most teams run every day. That is the calculation this article works through: the price, the benchmarks behind the claim, and what is still unproven.

What Gemini 4 Neon is

Neon is the fourth model in the Gemini 4 family, following Hydrogen, Argon and Carbon. Google describes it as bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work and science workflows. In the family lineup, Carbon takes the hardest long-horizon projects, Hydrogen handles fast everyday execution, and Neon balances depth and efficiency for complex, high-frequency work.

The model runs on a unified 1-million-token context window, the same size as its siblings, and supports long-decode generation up to 1 million tokens, according to Google's pricing documentation. That matters for the workloads Google is aiming at: multi-file software engineering, analytical reasoning across spreadsheets and PDFs, and agent workflows that run for a long time. Neon can synthesize spreadsheets, PDFs, charts, audio and video inside a single context window and turn them into structured deliverables.

Neon also gets four thinking levels, low, medium, high and xhigh, matching Hydrogen. Google says the levels scale steadily so teams can dial intelligence and token spend to what each workload needs. The default level is high, and most of the cost claims in the announcement are measured at that setting.

In the Gemini app, the launch reshuffles the model picker. According to 9to5Google, Neon replaces Argon, Gemini 3.1 Pro is going away, and the picker becomes Auto, Everyday help with Hydrogen, Deeper analysis with Neon, and Code and complex work with Carbon. Argon remains available as an API model, and its pricing page still lists it, so the replacement is an app change rather than a retirement of the model.

What it costs

Neon is priced at $3 per million input tokens, $0.15 per million cached input tokens, which Google describes as a 95% discount, and $15 per million output tokens. Those rates come from Google's own API pricing page, read on November 19, 2026. The table below puts Neon next to its siblings, with approximate UK and Canadian conversions at November 19, 2026 market rates of 1 USD = 0.7609 GBP and 1 USD = 1.4333 CAD. Google bills API usage in USD, so the pound and dollar figures are approximations.

ModelInput, per 1M tokensOutput, per 1M tokensCached input, per 1M tokensFree tier
Gemini 4 Neon$3.00 (£2.28 / CA$4.30)$15.00 (£11.41 / CA$21.50)$0.15 (£0.11 / CA$0.21)No
Gemini 4 Hydrogen$1.50 (£1.14 / CA$2.15)$7.50 (£5.71 / CA$10.75)$0.08 (£0.06 / CA$0.11)Yes
Gemini 4 Carbon (to Nov 30)$2.00 (£1.52 / CA$2.87)$10.00 (£7.61 / CA$14.33)$0.10 (£0.08 / CA$0.14)No
Gemini 4 Carbon (from Dec 1)$4.00 (£3.04 / CA$5.73)$20.00 (£15.22 / CA$28.67)$0.20 (£0.15 / CA$0.29)No
Gemini 4 Argon (to Nov 30)$2.00 (£1.52 / CA$2.87)$10.00 (£7.61 / CA$14.33)$0.10 (£0.08 / CA$0.14)No
Gemini 4 Argon (from Dec 1)$4.00 (£3.04 / CA$5.73)$20.00 (£15.22 / CA$28.67)$0.20 (£0.15 / CA$0.29)No

Three pieces of fine print matter before anyone budgets around these numbers. First, Neon is not available on the free tier, so every token is billed. Second, output pricing includes thinking tokens, and Google states that every token the model processes is counted and billed, including thinking tokens and internal tool system instructions when grounding is used. A long reasoning trace is not free. Third, context caching on Neon is implicit only. Explicit caching is not supported, so there is no way to pin a prompt prefix into cache manually and guarantee the $0.15 rate.

Grounding with Google Search is priced separately: 5,000 free search requests a month, shared across Gemini 3 and newer models, then $14 per 1,000 requests, subject to a daily cap of 1,500 requests. Grounding with Google Maps follows the same pattern, 5,000 free requests a month, then $14 per 1,000 search queries.

The December 1 flip

The pricing page confirms what our earlier Gemini 4 Argon coverage reported: Carbon and Argon cost $2 per million input tokens and $10 per million output tokens through November 30, 2026, then move to $4 and $20 on December 1. Cached input doubles at the same time, from $0.10 to $0.20.

That change reorders the family. Until November 30, Neon is the most expensive Gemini 4 model per token. From December 1, it is 25% cheaper than Carbon and Argon on both input and output, and its cached rate of $0.15 undercuts their $0.20. Hydrogen stays the cheapest at $1.50 and $7.50, exactly half of Neon's rates.

The practical question is whether the work you run actually lands on Neon rather than Carbon. Google still positions Carbon as the model for the hardest long-horizon projects, and Neon as the balance of depth and efficiency. If a task needs Carbon's ceiling, the December price rise is a cost increase with no cheaper substitute. If it runs fine on Neon, the same work gets 25% cheaper per token on December 1 without changing providers. Teams should test their heaviest prompts on both before the flip, while Carbon is still at its introductory rate.

The benchmarks, and how much to trust them

Google's announcement leads with a stack of benchmark results. The headline number is a score of 61 on the Artificial Analysis Intelligence Index at Neon's xhigh thinking level, ranking third overall, behind larger and slower models. Artificial Analysis is an independent benchmarking outfit, so that score is third-party measurement rather than a Google-run test, though Google chose to highlight it.

The coding numbers are the strongest part of the case. Neon ranks first on the Artificial Analysis Coding Agent Index with 58.6, scores 65.2% on Terminal-Bench 4.1, and reaches 46.2% on FrontierCode 2.0 at its default high setting. On SWE-sweep, it leads the mini harness at 12.8% and more than triples to 39.0% inside Google Antigravity, which shows what the Antigravity harness makes possible rather than what the raw model does on its own.

BenchmarkNeon resultWhat it measures
Artificial Analysis Intelligence Index61, ranked 3rd overall (xhigh)General intelligence across tasks
Artificial Analysis Coding Agent Index58.6, ranked 1stAgentic coding ability
Vals Index70.9%, ranked 2ndReal-world task performance
Terminal-Bench 4.165.2%Terminal-based coding tasks
FrontierCode 2.046.2% (high setting)Hard coding problems
Terminal-Bench Science70.5%, best in classScientific reasoning
RiemannBench84.0%, best in classMathematical reasoning
HLE-Diamond71.3%, best in classExpert-level questions
Agents' Last Exam v1.175.8%, on par with leading modelsLong-horizon agent work
Agent Arena Index1336, ranked 2ndReal-world agent sessions
Zapier AutomationBench56.3%, ranked 1stWorkflow automation
Andon Labs Vending-Bench 2$16,843, ranked 1stRunning a simulated business

Neon also takes first place on the Vals Legal Research Bench and Tax Agent Bench, scores 73.5% on Chartography for reading dense professional charts, and ranks first on most Surge AI benchmarks across finance, accounting, spreadsheet modeling and complex constraint following. On the Agent Arena Index it debuts second globally with 1336 across more than 4,200 real-world agent sessions, and ranks second on steerability, which measures how cleanly the model adapts to mid-task corrections.

Every figure above is Google-reported, and that is the caveat. The Artificial Analysis and Vals numbers come from independent leaderboards, but Google selected which results to publish, and no independent head-to-head of Neon against Carbon on identical prompts exists yet. The cost claims carry the same limit. Google says Neon costs 32% less per task than Carbon at the default high thinking level, and roughly 50% to 80% less per task than top competing frontier models, with xhigh costing 36% or more less per task than competing models at their highest setting. Those are Google's measurements on third-party benchmarks, and per-task cost depends on how much thinking a model does on your prompts. The per-token rates are verifiable arithmetic. The per-task figures are not, until someone else runs the same workloads.

The coding-model price war is moving fast. Microsoft's MAI Code 2 launched days earlier, claiming 98% of GPT-6.1 Sol's score at 40% of the price.

One framing deserves a correction before it hardens into received wisdom. Several reports describe Neon as Carbon quality at Flash speeds. That is a summary from 9to5Google, not a Google claim, and Google published no latency numbers for Neon. The blog says Neon delivers frontier-level reasoning faster and more cost efficiently than Carbon, without attaching a speed figure. Treat the speed part as plausible but unmeasured.

Neon for cyber, and who gets that version

Alongside the general model, Google announced Gemini 4 Neon for cyber, a variant tuned for defensive security work. It produces more secure code than Carbon across easy, medium and hard challenges on AutoBaxBench, reaching state of the art on the hard tier at 76%. On WebExploitBench v1.2 it scores 51.26% at half the cost of competing large models. On SEC-bench Pro it reaches 90.5% while costing roughly half as much per challenge, and on ExploitGym it reaches frontier-level performance using half the tokens Carbon needs. Patching is the strongest relative result: 70.7% on CWE-bench v2, a 3.4% uplift over Carbon.

This variant is not generally available. Google says Gemini 4 Neon for cyber is rolling out to trusted defenders in the coming days as part of the Fairwind Program, and access requests go through a Cloud account team. There is no published price for the cyber variant, and no self-serve route to it.

The safety section of the announcement is worth reading in full rather than summarized, because the numbers are unusually specific. Google says Neon maintains Carbon-level protections against severe misuse and prompt injection, refused 1.98% of tasks on Terminal-Bench 4.0 and 0% on Terminal-Bench-Science, matched Carbon with 0.0% cheating on Terminal-Bench 4.0, took shortcuts on 2.45% of tasks on its coding cheating evaluation, the lowest rate of any Gemini 4 model, and recorded a 0.09% misaligned action rate with a 99.2% pass rate on its agent honesty evaluation. On Gray Swan's indirect prompt injection benchmark, it resisted every attack at its highest thinking level. All of these are Google's own internal evaluations, and they should be read as the company's account of its own testing, not as independent verification.

How to get it

Neon began rolling out on November 18 in four places. Developers get it on the Gemini API paid tier in Google AI Studio, and in Android Studio. Enterprises get it in Gemini for business through the API or the model dropdown, and in Google Antigravity. Consumers get it in the Gemini app, but only as a paid subscriber.

The consumer route is the subscription, not the API. Google AI Pro costs $19.99 a month, or $199.99 a year, and includes 5 TB of storage. Google AI Ultra comes in two tiers: Ultra 5x at $99.99 a month with 20 TB of storage and five times Pro's Gemini usage limits, and Ultra 20x at $199.99 a month with 30 TB and twenty times Pro's limits. Those prices were verified on Google's plans page on November 19, 2026. The old single Ultra plan at $249.99 a month is gone. In the app, Neon sits under Deeper analysis in the model picker.

For developers, the API route is the one with a bill attached to usage rather than a flat subscription. The paid tier is the only tier: Neon is not available on the free tier, so a team evaluating it needs a billing-enabled project. Google AI Studio offers a way to try it with a paid Pro or Ultra subscription, and all Pro and Ultra subscribers can try it in AI Studio.

In Antigravity, Neon also plays a second role. Google says it works as a fast, high-judgment Orchestrator directing parallel Hydrogen subagents, cutting down on redundant file reads, sharing progress as it works, and handling context compaction in long sessions. That is a different job from answering a single prompt: Neon plans, Hydrogen executes, and the bill is mostly Hydrogen's cheaper tokens. It is the clearest sign of where Google thinks the model earns its keep, and it is the configuration worth testing first for teams already in Antigravity.

Common mistakes to avoid

First, do not compare Neon to Carbon at today's prices and assume the gap stays put. Until November 30, Carbon is cheaper per token than Neon. From December 1, Neon is 25% cheaper. Any cost comparison that does not state its date is wrong for one of the two periods.

Second, do not budget on the input rate alone. Output tokens cost five times input tokens, and thinking tokens are billed as output. A task that reasons at length can cost far more than the $3 input figure suggests, and the cached $0.15 rate only applies to implicitly cached input, not to output or thinking.

Third, do not assume the free tier covers a trial. Neon has no free tier. A team that wants to evaluate it needs either a paid API project or a Pro or Ultra subscription.

Fourth, do not treat the benchmark table as a buying decision. The scores are Google-reported, several come from Google's own harnesses, and the strongest coding result, 39.0% on SWE-sweep, was measured inside Antigravity rather than on the raw model. Run your own prompts on Neon and Carbon side by side before committing a workload to either.

Fifth, do not plan a security workflow around the cyber variant. It is rolling out to trusted defenders through the Fairwind Program, with no published price and no self-serve access. The general model is the one you can buy today.

Who should look at it

Neon fits teams that run coding and agent workloads all day and feel Carbon's bill. If your prompts are multi-file refactors, long agentic sessions, or analytical work across documents and spreadsheets, Neon's positioning is aimed directly at you, and the December 1 price flip makes the test cheap to run. It also fits Antigravity users, where the Orchestrator role gives Neon a job Hydrogen cannot do and Carbon does at a higher price.

It fits less well for the hardest long-horizon work, where Google still points to Carbon, and for casual use, where Hydrogen is half the price and available on the free tier. The honest test is a week of your own prompts on Neon at high thinking, measured against Carbon on the same tasks, before December 1 changes the arithmetic. If Neon holds up on your work, the flip turns a close call into a clear saving. If it does not, you will have learned that while Carbon is still at its introductory rate.

Is Gemini 4 Neon cheaper than Carbon?

From December 1, 2026, yes. Neon costs $3 per million input tokens and $15 per million output tokens, while Carbon rises to $4 and $20. Until November 30, Carbon is cheaper at its $2 and $10 introductory rates.

Can I use Gemini 4 Neon for free?

Not through the API. Google's pricing page states Neon is not available on the free tier, so every token is billed. In the Gemini app, Neon requires a Google AI Pro or Ultra subscription, starting at $19.99 a month.

How does Neon compare with Hydrogen?

Neon costs twice as much per token as Hydrogen, $3 versus $1.50 for input and $15 versus $7.50 for output, but targets harder work: complex coding, analytical reasoning and production-grade agent workflows, while Hydrogen handles fast everyday execution.

What is the context window on Gemini 4 Neon?

Neon has a unified 1-million-token context window, the same as the rest of the Gemini 4 family, with long-decode generation up to 1 million tokens, according to Google's pricing documentation.

Does Neon replace Argon?

In the Gemini app, yes. 9to5Google reports that Neon replaces Argon in the model picker and Gemini 3.1 Pro is going away. Argon remains available as an API model with its own pricing.

When does the Gemini 4 Neon cyber version launch?

There is no public launch date. Google says Gemini 4 Neon for cyber is rolling out to trusted defenders in the coming days through the Fairwind Program, with access arranged through a Cloud account team.

Gemini 4 Neon is the model Google expects most working developers to reach for, and the December 1 price flip is the reason to test it now. The benchmarks say it rivals Carbon where it counts. The bill, from December, will be 25% lower. Run your own prompts before the flip, and let the results, not the announcement, decide.

Sources

13 min read
57 views
edit_note Written by

Ratul Ahamed

Founder & Editor-in-Chief

Ratul Ahamed is the founder and editor of AI Core Updates, an independent site covering AI tools, models and pricing. He is a web developer based in Bangladesh and also runs All In One Tools, a free collection of browser-based utilities. Contact: [email protected]

Read all stories by Ratul Ahamed arrow_forward
verified_user Human-Reviewed Editorial

forum Comments (0)

Leave a Reply

chat_bubble_outline

No comments yet. Be the first to join the conversation!

Weekly AI Digest

Get our weekly AI tools digest

One email a week analyzing new AI tools, model updates, and practical workflows. Tested and summarized with zero fluff. Unsubscribe any time.

bolt
Hands-on AI Tool Reviews
Pricing math, limits, and alternatives.
verified_user
Strict Anti-Spam Policy
1-click unsubscribe at any time.
mark_email_unread

Subscribe to Digest

Sent weekly • No spam

person_outline
mail_outline

No spam ever. You can unsubscribe with a single click at any time.

cookie

Cookie Settings

We use cookies to enhance your browsing experience, serve personalized content, and analyze our traffic. By clicking "Accept All", you consent to our use of cookies. Learn more