cd /news/large-language-models/gemini-4-neon-3-15-pricing-1m-contex… · home › topics › large-language-models › article
[ARTICLE · art-172434] src=shattered.io ↗ pub= topic=large-language-models verified=true sentiment=· neutral

Gemini 4 Neon: $3/$15 Pricing, 1M Context, 4 Levels

Google released Gemini 4 Neon on November 18, 2026, as a generally available model in the Gemini API priced at $3 per million input tokens and $15 per million output tokens, with a 1-million-token context window and four thinking levels (low, medium, high as default, and xhigh). Google pitched the model, ID `gemini-4-neon`, for complex software engineering, knowledge work and science workflows, and made it available through Google AI Studio, Gemini for business, Google Antigravity and the Gemini app for paid Google AI Pro and Ultra subscribers. TheDecoder reported Neon as Google's fourth new Gemini model in a month, following Hydrogen, Carbon and Argon, a count the outlet supplied rather than Google.

key takeaway

Gemini 4 Neon is generally available at $3 per 1M input tokens, $0.15 per 1M cached input tokens and $15 per 1M output tokens, with a 1M-token context window and low/medium/high/xhigh thinking levels.

by read20 min views3 publishedNov 19, 2026
Gemini 4 Neon: $3/$15 Pricing, 1M Context, 4 Levels
Image: Shattered (auto-discovered)

Google released Gemini 4 Neon on November 18, 2026, and the company’s launch post sells it as “frontier-level intelligence with fast end-to-end task completion” for software engineering, knowledge work, and science. The model ID is gemini-4-neon. It costs $3 per million input tokens and $15 per million output tokens, and it carries a 1-million-token context window.

TheDecoder framed the release as Google’s fourth new Gemini model in a month. That count comes from the outlet, not from Google’s own announcement, so read it as reporting rather than an official tally. What nobody disputes is the pace. Hydrogen, Carbon, and Argon all landed before Neon, and our earlier piece on Gemini 4 Hydrogen’s $1.50 input price and 1M-token window shows how fast the family has filled out.

This analysis covers what Google confirmed, what only third parties claim, and what Gemini 4 Neon pricing means for teams budgeting agent workloads. It also sorts the four-model lineup, runs the cost math against Hydrogen and Carbon, and ends with five predictions. Where the official materials stay silent, this article says so instead of guessing.

Don't miss new tech stories on Google

Add Shattered once in the Google app and our stories show up in your news suggestions.

Add now

What Google Announced on November 18 #

The core announcement is short. Google named a new model, Gemini 4 Neon, gave it the ID gemini-4-neon, and put a price on it the same day. The Gemini API release notes mark the model as generally available on November 18, 2026, which means developers did not have to wait through a preview period to ship against it.

Google describes the intended work as complex software engineering, knowledge work, and science workflows. That is a narrower pitch than a general chat model. It tells you who the target buyer is: teams running long, multi-step jobs where speed and cost per task matter as much as raw quality.

The rollout spans several surfaces at once. Neon is live in the Gemini API through Google AI Studio, in Gemini for business, in Google Antigravity, and in the Gemini app for eligible paid Google AI Pro and Ultra subscribers. A same-day launch across the API, the enterprise product, the coding environment, and the consumer app is a sign Google wants Neon to be the workhorse tier, not a lab showpiece.

Two things Google did not put in the headline are worth flagging early. The launch materials we reviewed do not state that Neon is the fourth Gemini model in a month, and they do not claim Neon matches or beats Gemini 4 Carbon. Both ideas circulate in third-party coverage, and both get a closer look below.

Gemini 4 Neon Specs at a Glance #

Specs are where a launch either holds up or falls apart. The table below lists only what Google’s materials and named outlets confirm. Anything missing from the sources, such as official benchmark scores, is left out on purpose.

Spec Gemini 4 Neon
Model ID gemini-4-neon
Announced November 18, 2026
Status Generally available (GA) in the Gemini API from November 18, 2026
Context window 1 million tokens
Maximum decode Up to 1 million tokens
Thinking levels low, medium, high (default), xhigh
Input price $3 per 1M tokens
Cached input price $0.15 per 1M tokens
Output price $15 per 1M tokens
Inputs handled (per TheDecoder) Text, images, video, audio, PDFs
Not supported (per TheDecoder) Image and audio generation

Two details stand out. First, a 1-million-token window paired with up to 1 million tokens of output is a notable pairing for long jobs. A long decode matters for jobs such as whole-codebase refactors or long research briefs, where a model needs room to write as well as read. Second, the four thinking levels give developers a dial instead of a fixed behavior.

TheDecoder adds that Neon supports Computer Use, code execution, file search, and grounding through Google Search and Maps. Those are the building blocks of agent workflows, and they line up with the software engineering and knowledge work focus in Google’s own description.

Gemini 4 Neon Pricing: $3, $0.15, and $15 Per Million Tokens #

Google set Gemini 4 Neon pricing at $3 per million input tokens, $0.15 per million cached input tokens, and $15 per million output tokens. Output costs exactly five times input, which is the ratio that decides the bill for any workload with long generations. A coding agent that writes 30,000 tokens per task feels the $15 line far more than the $3 line.

Against its siblings, Neon sits in the middle of the range. TheDecoder lists Hydrogen at $1.50 and $7.50, so Neon costs twice as much per token on both sides. Carbon and Argon carry an introductory rate of $2 input and $10 output, which is cheaper than Neon today, but TheDecoder reports that rate doubles to $4 and $20 on December 1. Once that step lands, Neon’s $3 and $15 undercut both by 25 percent.

What the $0.15 cached rate means

The cached input price is 5 percent of the standard input price, a 95 percent discount. For agent loops that resend the same long system prompt, tool definitions, or codebase context on every turn, that line can matter more than the headline rate. A team that caches 300,000 of a 400,000-token prompt pays $0.045 for the cached part instead of $0.90.

The caveat is that caching only helps when prompts repeat. One-off questions with fresh context pay the full $3. Teams should measure their real cache hit rate before assuming the discount shows up on the invoice.

The Gemini 4 Lineup After Neon: Hydrogen, Carbon, Argon #

Neon arrives into a family that is only a few weeks old. According to 9to5Google, the Gemini 4 family before Neon consisted of three models. Hydrogen handles everyday help and acts as a replacement for the old Flash tier. Argon targets code and analysis. Carbon is meant for the most complex work. Neon adds a fourth lane between them on price and speed.

Model Input per 1M tokens Output per 1M tokens Positioning (per 9to5Google)
Gemini 4 Hydrogen $1.50 $7.50 Everyday help, Flash replacement
Gemini 4 Argon $2 intro, then $4 $10 intro, then $20 Code and analysis
Gemini 4 Carbon $2 intro, then $4 $10 intro, then $20 Most complex work
Gemini 4 Neon $3 $15 Frontier-level reasoning for faster, cheaper high-frequency workflows

Prices come from TheDecoder’s launch coverage. The intro-rate step is the detail to watch, and we covered the mechanics in our report on Carbon’s price doubling to $4 after November 30. Neon launched at a flat rate with no promotional window, which makes budgeting simpler.

The practical read is a three-step ladder. Hydrogen is the cheap, fast tier. Neon is the balanced tier. Carbon is the depth tier. Argon’s place is less clear, because 9to5Google reports that Neon replaces it in the Gemini app. Google has not said in the materials cited here what happens to Argon in the API.

Is Neon Really Google’s Fourth New Gemini Model in a Month? #

The headline claim deserves a straight answer. TheDecoder wrote that Neon is Google’s fourth new Gemini 4 model in four weeks. Google’s official announcement does not say this, and the fact sheet behind this article marks the “fourth in a month” count as unconfirmed by the cited official materials.

The count is plausible, though. Three named models stood in the Gemini 4 family before Neon, according to 9to5Google, and Neon makes four. The outlet also says Carbon and Hydrogen shipped the week before Neon. If Argon arrived within the same four-week window, TheDecoder’s math works out. Google has not published a release calendar that settles the question.

Why does the count matter? Because release cadence is a competitive signal. Four models in roughly a month means Google is shipping tiers faster than most buyers can evaluate them. That helps developers who want options and hurts procurement teams that need a stable choice to standardize on. A new model every week or two makes any internal benchmark go stale quickly.

Our position: cite the four-in-a-month figure as TheDecoder’s framing, and anchor any hard claims to Google’s own pages. That is the approach used throughout this piece.

Does Neon Match Gemini 4 Carbon? What Is Confirmed #

9to5Google’s headline says Neon brings Carbon quality at Flash speeds, and its report states that Neon performs comparably to Carbon across coding, reasoning, and agentic tasks, possibly better. That is a third-party characterization. The cited Google announcement does not make a Neon-versus-Carbon claim, and this article does not treat it as settled.

The distinction is not pedantry. Parity claims are the easiest thing to repeat and the hardest to verify. They depend on which benchmarks ran, which thinking level was set, and how many tokens each model burned per answer. TheDecoder noted that benchmarks should follow within hours of the launch, but we have no verified score tables to report as of this writing, so we publish none.

Here is what we can say with confidence. Google priced Neon below Carbon’s post-promotion rate, positioned it for high-frequency workflows, and put it in front of Pro subscribers in the Gemini app. Those choices suggest Google sees Neon as the model most people should use most of the time. They do not prove Neon is as smart as Carbon on the hardest tasks.

For buyers, the right move is a head-to-head on your own tasks. Run the same 50 or 100 representative prompts through Neon at high and Carbon at its default, then compare accuracy, latency, and total tokens. Our walkthrough of calling Gemini 4 Carbon through the API in 12 steps gives a setup you can adapt for Neon by swapping the model ID.

Thinking Levels: low, medium, high, and xhigh #

Neon exposes four thinking levels: low, medium, high, and xhigh. The default is high. That default is a quiet statement of intent. Google chose to ship Neon with deep reasoning turned on, then let developers dial it down for speed instead of dialing it up for quality.

The levels change the cost-quality trade on every request. A low setting suits classification, extraction, and routing, where answers are short and latency matters. A medium setting fits drafting and summarization. The high and xhigh settings belong to debugging, multi-file edits, and research tasks where a wrong answer costs more than a few extra seconds.

The open question is token spend. Thinking tokens are typically billed alongside output, so a higher level can raise the bill even though the per-token rate stays at $15. The launch materials cited here do not quantify how many extra tokens each level uses. Until Google or independent testers publish that data, assume xhigh costs meaningfully more per task than low, and measure it.

A sensible starting policy for engineering teams is simple. Route cheap, high-volume calls to low or to Hydrogen. Send normal agent turns to Neon at medium or high. Reserve xhigh and Carbon for the 5 or 10 percent of tasks that fail at lower settings. Then track the fallback rate weekly.

Where Developers Can Use Gemini 4 Neon Today #

Availability is broad on day one. The model runs in the Gemini API through Google AI Studio, in Gemini for business, in the Gemini app for eligible paid Google AI Pro and Ultra subscribers, and in Google Antigravity. Each surface serves a different buyer, so the rollout reads like a coordinated push and not a staggered one.

For coding teams, Antigravity is the most interesting entry. Google Antigravity’s blog post on Neon covers the model’s arrival in the agentic coding environment, and paid Pro and Ultra subscribers get access there. Putting a speed-focused model inside an agent-first IDE makes sense, since agents make many calls per task and punish slow or expensive models.

For enterprises, Gemini for business gives Neon a place in the API and in the model dropdown. That matters because many companies cannot adopt a model until their admin console exposes it. Day-one inclusion removes one common source of delay.

For hobbyists and prototypers, AI Studio is the quickest path. Developers can swap gemini-4-neon into an existing Gemini API call and compare results against older models without changing the rest of their code. That low switching cost is part of Google’s strategy, and it is also why model churn is easy for developers to absorb.

What Neon Changes in the Gemini App #

Consumers see the change in the model picker. 9to5Google reports that the picker now offers Auto, Everyday help on Hydrogen, Deeper analysis on Neon, and Code and complex work on Carbon. Neon is open to AI Pro and AI Ultra subscribers, and it replaces Argon in the app.

The same report says Gemini 3.1 Pro is going away in the app, and that 3.5 Flash-Lite now sits behind the Auto option for paid users. In plain terms, Google is retiring the older Pro-and-Flash naming in the consumer product and replacing it with task labels. The model names still show, but the labels do the steering.

According to 9to5Google, Google says Neon in the app should write more clearly, read uploads like spreadsheets, PDFs, and photos more precisely, and pull in details from Gmail and Docs. It also says Tasks in Gemini can now run on Neon, which should make multi-step jobs faster and stretch usage limits further. Those are Google’s expectations as relayed by the outlet, not measured results.

The business logic is plain. Putting the mid-priced model in front of Pro subscribers lets Google serve most requests on a cheaper-to-run model than Carbon while still advertising frontier-level reasoning. That likely helps Google’s margins, and it fits with Neon, not Carbon, taking the “deeper analysis” slot.

Cost Math: Neon Against Hydrogen and Carbon #

Per-token prices hide the number that finance teams care about, which is cost per finished task. The table below runs an illustrative agent task with 400,000 input tokens and 30,000 output tokens. This is arithmetic on published rates, not a measured benchmark. Real tasks differ, and models differ in how many tokens they spend to reach an answer.

Scenario Cost per task Cost per 1,000 tasks
Hydrogen ($1.50 / $7.50) $0.825 $825
Carbon or Argon, intro rate ($2 / $10) $1.10 $1,100
Neon, no caching ($3 / $15) $1.65 $1,650
Neon, 300,000 input tokens cached $0.795 $795
Carbon or Argon, after intro ($4 / $20) $2.20 $2,200

The cached row is the surprise. With 75 percent of the prompt cached, Neon comes in under Hydrogen’s uncached cost for the same task. We do not list cached rates for the other models because the cited sources do not give them, so that comparison is not apples to apples.

A quick calculator you can adapt

RATES = {  # USD per 1M tokens: (input, cached input, output)
    "gemini-4-neon": (3.00, 0.15, 15.00),
}

def task_cost(model, fresh_in, cached_in, out):
    inp, cached, outp = RATES[model]
    return (fresh_in * inp + cached_in * cached + out * outp) / 1_000_000

print(task_cost("gemini-4-neon", 400_000, 0, 30_000))        # 1.65
print(task_cost("gemini-4-neon", 100_000, 300_000, 30_000))  # 0.795

Swap in your own token counts from API usage logs. For a fuller picture of how cost per task can diverge from price per token across vendors, see our earlier comparison of Gemini Hydrogen’s task cost against Claude’s.

How Neon Stacks Up Against Rival Frontier Models #

Google is not alone in tiering its flagship line. TheDecoder compared the Gemini 4 lineup with the rivals’ by suggesting Carbon and Argon are Google’s counterparts to OpenAI’s Astra and Anthropic’s Fable, and that Neon is likely its answer to Claude Opus or Sol. That is the outlet’s framing. Neither Google nor the rivals have drawn the comparison themselves in the materials we reviewed.

We are not building a rival price table here, because the sources cited for this article carry no verified prices or benchmark scores for OpenAI or Anthropic models. Putting numbers in a table without sourcing them would break the standard this piece set at the start. For OpenAI’s side of the story, our report on GPT-6.2 Astra’s claimed 5.2-point gain at one-third the cost covers the vendor’s own claims.

What the three vendors share is a strategy. Each now sells at least three rungs, a cheap fast model, a balanced one, and a deep one. The competition has moved from who has the single best model to who offers the best price at each rung. Neon is Google’s bid for the middle rung, where most production traffic lives.

The distinguishing features on Google’s side are the 1-million-token window, the 1-million-token decode, and the same-day reach across the API, Antigravity, the enterprise product, and the consumer app. Buyers should weigh those against their own needs, such as data residency, existing cloud contracts, and tool ecosystems, none of which a launch post resolves.

Historical Context: From Flash and Pro to Four Element Names #

Google’s older lineup used simple labels. Flash meant fast and cheap, Pro meant capable and expensive. The 9to5Google report still references Gemini 3.1 Pro and 3.5 Flash-Lite as the app’s outgoing options, which shows how recently that scheme was in use. Hydrogen now plays the Flash role, and the other three names cover the rungs above it.

The new scheme trades clarity for room. Flash and Pro told you the rank at a glance. Hydrogen, Argon, Carbon, and Neon tell you nothing without a chart. 9to5Google’s launch coverage even shows readers complaining that the names are hard to keep straight. The app’s task labels partly offset that by mapping each model to a use.

The speed of the build-out is the bigger historical point. Our earlier reporting traced Gemini 4 from the Argon agent safety episode, covered in our report on Gemini 4 Argon agents reaching a live Epoch AI site, through the Hydrogen and Carbon launches and now to Neon. That is a lot of product surface in a short window, and it follows a broader pattern across AI labs in 2026 of shipping tiers rather than single flagships. Our AI and machine learning coverage tracks the rest of that field.

Market Impact: Why the Middle Tier Matters #

Pricing power in AI sits in the middle tier. Cheap models win on volume, deep models win on headlines, and balanced models win on revenue because teams run them all day. By pricing Neon at $3 and $15, Google has set an anchor that sits between its own cheap and deep tiers, and 25 percent under the post-promotion rate of Carbon and Argon.

For developers, the near-term effect is a routing question. Teams that sent everything to one premium model can now split traffic across three Google tiers and lower their bills without rewriting prompts. For vendors, the effect is pressure. If a frontier-class model at $3 and $15 holds up in testing, rivals face a harder case for charging more for similar quality.

For Google, the strategic payoff is lock-in through convenience. A single Gemini API key reaches four tiers, with the same tools, the same context window, and the same thinking-level dial. Moving a workload between tiers takes one string change. Moving it to another vendor takes a re-evaluation.

The risk is confusion. Four closely priced models with overlapping roles can slow decisions, and 9to5Google’s reader comments show the naming already irritates some users. Google also has an unresolved question in the API about Argon, since Neon takes its slot in the app. Teams building on Argon should ask their Google contacts about its roadmap.

Five Predictions for Gemini 4 After Neon #

These are our own forecasts, not reported facts. They rest on the pricing, availability, and positioning described above.

  1. Benchmark tables settle the Carbon debate within weeks. Independent testers will publish Neon-versus-Carbon results soon, and the answer will probably vary by task. Expect Neon to look strong on speed-adjusted measures and Carbon to hold an edge on the hardest problems.
  2. The December 1 price step pushes teams toward Neon. Once Carbon and Argon rise to $4 and $20, Neon’s flat $3 and $15 becomes the obvious default for anyone who does not need maximum depth.
  3. Antigravity makes Neon its everyday coding model. Agent-heavy products reward speed and low cost, so Neon is a natural default there. Hydrogen handles light calls and Carbon handles the hard ones.
  4. Google clarifies Argon’s future. With Neon replacing Argon in the app, a statement on Argon’s API status is likely within a quarter.
  5. Rivals respond with tier or price moves. A strong mid-priced Google model invites a reply from OpenAI and Anthropic, probably as new promotional rates or a new middle tier before spring.

Any of these could miss. The first is the most testable, since the data should arrive fast.

What Teams Should Do This Week #

A launch this size calls for a short, disciplined evaluation, not a rushed migration. Start by writing down your three most expensive production workloads and their current model, token counts, and monthly cost. Without that baseline, no pricing comparison means anything.

Next, replay a fixed set of real prompts through gemini-4-neon at medium and high. Record accuracy, latency, and total tokens, then compute cost per successful task. If your prompts share a long prefix, test with caching on, because the $0.15 rate changes the math.

Then decide on routing. Keep cheap calls on Hydrogen, put the bulk on Neon if it passes, and keep Carbon as the escalation path. Review the split again after December 1, when Carbon and Argon leave their introductory rates. Finally, treat every parity claim, including 9to5Google’s Carbon comparison, as a hypothesis to test on your own data.

FAQ: Gemini 4 Neon #

What is Gemini 4 Neon? It is a Google model with the ID gemini-4-neon, announced November 18, 2026. Google describes it as delivering frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows.

How much does Gemini 4 Neon cost? The Gemini API price is $3 per 1 million input tokens, $0.15 per 1 million cached input tokens, and $15 per 1 million output tokens.

What is the Gemini 4 Neon context window? The model has a 1-million-token context window and can decode up to 1 million tokens of output.

Is Gemini 4 Neon better than Gemini 4 Carbon? Google’s cited announcement does not say so. 9to5Google reports comparable or better performance across coding, reasoning, and agentic tasks, but that is third-party reporting. Test both on your own workload before choosing.

Where can I use Gemini 4 Neon? It is available in the Gemini API through Google AI Studio, in Gemini for business, in Google Antigravity, and in the Gemini app for eligible paid Google AI Pro and Ultra subscribers.

What thinking levels does Neon support? It supports low, medium, high, and xhigh. The default is high.

Is Neon really Google’s fourth new Gemini model in a month? TheDecoder reported it that way. Google’s official materials do not state the count, so treat it as the outlet’s framing, not a confirmed figure.

Can Gemini 4 Neon generate images or audio? According to TheDecoder, no. Neon accepts text, images, video, audio, and PDFs as input, but it does not generate images or audio.

── more in #large-language-models 4 stories · sorted by recency
── more on @google 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/gemini-4-neon-3-15-p…] indexed:0 read:20min 2026-11-19 · —