deep-dive

Gemini 4 Neon Explained: Specs, Pricing, and Benchmarks

The article explains that Gemini 4 Neon is Google's mid-tier model launched November 18, 2026 between Hydrogen and Carbon, replacing Argon as the Deeper analysis option. It details unconfirmed pricing, positioning as Carbon-level reasoning at Flash speed, independent benchmark results on Intelligence, Coding Agent, Vals and Agent Arena indexes, and access via Gemini app for Pro/Ultra, Antigravity, and Vertex AI, with guidance on when to choose Neon over Carbon.

November 21, 2026
·
9
min read
A man works at a wooden desk, using a laptop and desktop computer displaying code. A notebook, pen, mug, and small plants are also on the desk.

What Is Gemini 4 Neon?

Gemini 4 Neon is Google's November 18, 2026 addition to the Gemini 4 family, positioned between Hydrogen and Carbon as a model that delivers frontier-level reasoning for complex, high-frequency coding and analytical workflows at Flash-class speed. It replaces Argon in the Gemini app picker, and the confirmed Gemini 4 lineup is now Hydrogen, Neon, and Carbon, as described in Google's official announcement.

That correction matters because early summaries reduced the Gemini 4 family to Neon and Hydrogen. Google's own page positions Neon explicitly alongside Carbon and Hydrogen: while Carbon tackles the hardest, long-horizon projects and Hydrogen handles fast everyday execution, Neon balances depth and efficiency, with performance comparable to, if not surpassing, Carbon at Flash speeds and lower cost.

Independent coverage of the full lineup clarifies the rest of the puzzle. Before November 18, the Gemini 4 family consisted of Hydrogen for everyday help as the Flash replacement, Argon for code and analysis, and Carbon for most complex work. With this launch, Google places Neon between Hydrogen and Carbon, and Neon takes over Argon's slot in the Gemini app model picker as the "Deeper analysis" option.

Neon is positioned as Carbon-quality intelligence optimized for sustained, high-frequency coding and analytical work, rather than a lighter Hydrogen variant. With the lineup correctly placed, the next question is how Neon actually prices against its siblings.

Gemini 4 Neon vs. Argon and Carbon: Pricing and Positioning

Google Cloud's Model Garden pricing page lists Hydrogen, Argon, Neon, and Carbon together as the confirmed Gemini 4 family, but a full side-by-side rate card for Neon and Hydrogen was not available to verify at the time of writing. Treat per-token pricing for those two as unconfirmed until the provider page shows a dedicated row for each. What is confirmed is the structure around Argon and Carbon: both carry a promotional rate through November 30, 2026, stepping up to a higher steady-state rate starting December 1, 2026, while Neon is reported to carry no promotional window of its own.

That absence of a promo is why early coverage frames Neon as cheaper than Carbon once Carbon's promotional period ends, but confirming the exact gap requires checking the live pricing page directly rather than assuming a fixed ratio. Carbon and Argon's promotional rates may currently undercut Neon until those promotions expire.

Model Core focus Speed tier
Gemini 4 Neon Carbon-level reasoning at Flash speeds, agentic coding Flash-speed
Gemini 4 Argon (from Dec 1, 2026) Balanced production, cost-efficient reasoning Fast
Gemini 4 Carbon (from Dec 1, 2026) Maximum reasoning for hardest long-horizon tasks Slowest / most capable
Gemini 4 Hydrogen Everyday assistant, entry tier Fastest

The trade-off for a buyer is straightforward when you read the table with that December 1 date in mind.

Hydrogen is the low-cost entry point, positioned by Google as the everyday assistant tier in the Gemini app.

Argon and Carbon, once their steady-state rates take effect in December, are the performance bookends. Carbon is described in current coverage as the model for the hardest long-horizon problems, so you pay for depth and persistence. Argon maps to balanced production use where you want more reasoning than Hydrogen without Carbon's cost.

Neon is Google's attempt to collapse that choice. It is marketed as Carbon-level reasoning at Flash speeds, aimed at agentic coding loops and multimodal builds where you would have picked Carbon for quality but needed Flash-like latency. It is positioned as the mid-premium workhorse, not the cheapest to run, not the heaviest reasoner, but the option built for speed-constrained reasoning tasks where token volume is high and task cost matters more than raw per-token rate.

How Gemini 4 Neon Ranks on Independent Benchmarks

Gemini 4 Neon scores 61 at xhigh on the Artificial Analysis Intelligence Index as of Nov 18, 2026, and per Google's announcement ranks #3 overall on that index, #1 on the Artificial Analysis Coding Agent Index at 58.6, and #2 on both the Vals Index at 70.9% and the Agent Arena Index at 1336.

An open laptop displays a performance overview graphic with bar charts, line graphs, and a pie chart, accompanied by a notebook, pen, and a plant.
Independent benchmark scores help decide when to route agent traffic to Neon.
Turn Your Blog into a Sustainable Organic Demand Engine with Structured Content

HarperFlow publishes highly structured articles featuring FAQs, data tables, and direct-answer blocks that meet the rigorous citation standards required by AI search engines. By continuously auditing and improving your content through AI answer analytics, HarperFlow helps your site build long-term authority and visibility that outlasts ad-dependent strategies.

Start Your Trial Today →

Deciding whether to route production agent traffic through Neon this week means weighing those independent leaderboards against Google's blog claims. Each measures something different, and each was updated within hours of the Nov 18 launch, which is why checking the provider site directly still matters.

Artificial Analysis Intelligence Index

The Intelligence Index composite in v5.0.1 averages nine external evaluations: AA-Briefcase v1.1, AutomationBench-AA, Terminal-Bench 4.0, FrontierCode-AA 2.0, AA-Omniscience v1.1, GDP.pdf, HLE-Diamond, CritPt v2 (Hard Subset), and Terminal-Bench-Science 0.1. It is meant to capture general frontier reasoning across coding, spreadsheets, science, and long-horizon tool use.

Google's post reports Neon at 61 xhigh for #3 overall, just behind larger, slower models. The provider page confirms the 61 score for Neon Xhigh and lists the nine evaluations, but does not surface an explicit "#3 overall" rank in the text fetched for this piece. Treat the 61 as independently confirmed, the overall rank as a Google-attributed claim until the provider leaderboard shows it directly.

Artificial Analysis Coding Agent Index

This companion index isolates real agentic coding performance rather than Q&A. Google reports Neon at #1 overall with 58.6, citing 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at its default high setting. Those figures come from Google's announcement at time of writing; a separate coding-agent leaderboard page could not be pulled in this window, so treat the ranking as vendor-reported pending a direct check on artificialanalysis.ai/agents.

Vals Index

The Vals Index scores models on enterprise knowledge work. Vals builds separate benches for tasks like Legal Research and Tax Agent, then rolls them into a main index where higher is better. Google reports Neon at #2 with 70.9%, plus #1 on Legal Research Bench and Tax Agent Bench. That #2 rank and 70.9% were not independently fetched from vals.ai in this run, only via Google, so treat them as reported rather than verified on the provider.

Agent Arena Index

The Agent Arena Index from Arena.ai rates agents on live user sessions, not static questions. It blends task completion, praise versus complaints, handling of mid-task corrections (steerability), and Bash-command reliability across a large set of real sessions. Google reports Neon debuting at #2 globally with a 1336 score and #2 on steerability. Third-party coverage also noted the same #2 position at 1336. Direct confirmation still requires loading the Arena leaderboard itself, since the model is hours old at the time of this writing.

Same-day benchmark ranks for a new launch should be checked on the provider's own leaderboard before you cite them as settled fact.

Fast-moving AI news calls for sourcing discipline: only verifiable, dated facts earn a place in a comparison table.

Where to Access Gemini 4 Neon: API, Antigravity CLI, and AI Pro

Gemini 4 Neon is accessible as of November 18, 2026, through three live paths: the Gemini app for paid AI Pro and AI Ultra subscribers, Google Antigravity across app, CLI, and IDE extensions, and the Vertex AI model garden for developers. The mechanics differ by surface, and availability is not automatic everywhere.

Gemini app for AI Pro and Ultra

Google's reported picker layout places Neon as the "Deeper analysis" option, taking over the slot previously held by Argon, while 3.1 Pro is said to be removed and Auto maps to 3.5 Flash-Lite for paid users, a configuration described in secondary coverage rather than confirmed directly on Google's own page. 9to5Google reports Neon is live for AI Pro and AI Ultra subscribers in the Gemini app, and the same report notes availability in Google Antigravity for paid AI Pro or Ultra subscribers. That live status applies only to paid tiers, not to free accounts, and reflects the November 18 announcement date.

Antigravity app, CLI, and IDE extensions

For code-focused work, Google's Antigravity team states Neon is rolling out today in the Antigravity app, Antigravity CLI, and Antigravity IDE Extensions. In that surface Neon offers a context window of up to 512K tokens and four thinking levels: Low, Medium, High as default, and Extra High, with CLI flags --model gemini-4-neon-xhigh or --effort xhigh.

Tier gating is explicit. Free and Google AI Plus do not include Neon or Orchestrator (Neon). Paid AI Pro and Ultra include both. Trial Pro only includes them on trials tied to eligible premium devices. Enterprise includes Neon through Gemini for business but does not include Orchestrator (Neon). Orchestrator (Neon) itself pairs Neon for planning and review with Hydrogen for implementation, handing control back to Neon if Hydrogen gets stuck.

Vertex AI and Gemini API for developers

For production use, Neon appears in the Vertex AI model garden on the global endpoint. Because official model-version documentation has lagged behind SKU appearance during this launch window, check your own project console and release notes before adding Neon to an automated router. If your org uses routing logic that auto-selects models, treat Neon as confirmed only once it appears in your allow-list.

A release that moves this fast demands dated checks on each surface: publish date, primary source, and tier noted for every access claim.

Should You Build on Gemini 4 Neon Right Now?

Pick Gemini 4 Neon as the default build target if your workload is what Google described as frontier-level reasoning for complex, high-frequency workflows: production-grade software engineering agents sustaining long, tool-heavy sessions, enterprise analysis across spreadsheets, PDFs and charts in a large context window, and automation tasks where steerability and multi-step instruction following matter more than absolute ceiling intelligence.

For paid Pro and Ultra subscribers and for developers already calling the Gemini API paid tier, Neon becomes the cost-efficient workhorse. It was built to deliver comparable results to Carbon across coding, reasoning and agentic benchmarks while running at Flash-class speed, which reduces cost per completed task compared with larger frontier models. Automation only pays off here if the quality holds, and early benchmark reporting suggests it does.

Do not pick Neon if you need Carbon's absolute ceiling. Google still positions Carbon for the hardest, long-horizon projects that test the limits of intelligence, and for those, Carbon remains the better fit. Also hold off if you are on free, Plus, or other tiers where Neon has not rolled out yet, since early coverage shows Neon live for Pro and Ultra only, with Hydrogen remaining the default for everyday help. If you are still on Argon's promotional API pricing until November 30, test Neon side-by-side before switching, because short-term price parity can flip the cost math until the promo ends.

There are genuine open questions to log as unresolved rather than guess at. Google has not published a full regional rollout schedule, a firm long-term pricing commitment beyond the post-November 30 change, or a deprecation date for Argon on the API. Availability through Vertex AI model garden updates and enterprise entitlements is also still filtering through.

Teams shipping code agents, analytical assistants, or business automation today who already pay for Pro, Ultra, or API access can standardize on Neon now and reserve Carbon for the hardest evals, a split that gives speed and budget headroom without giving up frontier quality.

Sources

  1. Introducing Gemini 4 Neon
  2. Google launches Gemini 4 Neon with Carbon quality at Flash speeds, live for AI Pro
  3. Agent Platform Pricing
  4. Gemini 4 Neon Models - Intelligence, Performance & Price Comparison | Artificial Analysis
  5. Gemini 4 Neon and Orchestrator (Neon) are now available in Google Antigravity - Google Antigravity - Google AI Developers Forum

Frequently Asked Questions

Do I need a paid plan to use Gemini 4 Neon in the Gemini app?

Yes. Neon is live for AI Pro and AI Ultra subscribers as the Deeper analysis option, while Free accounts do not get it. The same report notes availability in Google Antigravity for paid AI Pro or Ultra subscribers and the change took effect on Nov 18, 2026.

How does Neon actually compare to Carbon and Hydrogen in daily use?

Carbon is positioned for your hardest, long-horizon projects and Hydrogen for fast everyday execution, while Neon balances depth and efficiency. Neon is described as bringing performance comparable to, if not surpassing, Carbon but at Flash speeds and lower cost. Choose Carbon for ceiling intelligence, Neon for sustained high-frequency coding, Hydrogen for quick tasks.

How large is Neon’s context window and where does it apply?

In Antigravity, Neon offers up to 512K tokens of context, which helps with large codebases and long documents. That limit is confirmed for the Antigravity app, CLI, and IDE extensions. Check your Vertex AI console for limits on other surfaces.

How do I select Neon when using the Antigravity CLI?

Use the flags --model gemini-4-neon-xhigh or --effort xhigh in the Antigravity CLI. Both were listed in the Antigravity rollout note for the launch. The default reasoning level is High, with Low, Medium, and Extra also available.

What thinking levels does Neon offer and which is default?

Neon exposes four levels: Low, Medium, High (default), and Extra. High is intended as the balanced default for most production work. Use Extra when you need maximum reasoning and can accept slower responses.

Does Neon fully replace Argon across all Google surfaces?

Neon replaces Argon in the Gemini app picker as the Deeper analysis option. Google Cloud's Model Garden pricing page still lists both Argon and Neon as separate SKUs. Treat the replacement as app-specific until docs remove Argon from API listings.

How should I think about Neon pricing versus Carbon’s promotional price?

Carbon’s confirmed input price is $2.00 per 1M tokens through Nov 30, 2026 on the global endpoint. Argon and Carbon step up on Dec 1, while Neon has no reported promo window. Until the promo ends, Carbon can undercut Neon, so test side by side before switching.

Which independent benchmarks support Neon’s coding claims?

Google reported Neon at 61 on the Artificial Analysis Intelligence Index at xhigh for #3 overall, #1 on Coding Agent Index at 58.6, #2 on Vals Index at 70.9% and #2 on Agent Arena Index at 1336. The Intelligence Index v5.0.1 averages nine evaluations including Terminal-Bench 4.0 and FrontierCode-AA 2.0. Verify ranks directly on provider leaderboards.

Turn Your Blog into a Sustainable Organic Demand Engine with Structured Content

HarperFlow publishes highly structured articles featuring FAQs, data tables, and direct-answer blocks that meet the rigorous citation standards required by AI search engines. By continuously auditing and improving your content through AI answer analytics, HarperFlow helps your site build long-term authority and visibility that outlasts ad-dependent strategies.

Start Your Trial Today
Written by
Hesham Mashhour
Founder @HarperFlow

Lover of all things automation and all things content.