Gemini 4 Neon: pricing, benchmarks, and Neon vs Carbon, Argon and Hydrogen

Gemini 4 Neon costs $3/$15 per million tokens, has a 1M context window and no free tier. Here's how it compares with Carbon, Argon and Hydrogen.

A glowing orange neon tube floating above a laptop keyboard at night, with blurred code on the screen behind it
Illustration generated for Late Night Softwares

TL;DR: Google released Gemini 4 Neon on 18 November 2026. It costs $3 per million input tokens and $15 per million output tokens, has a 1M-token context window, and isn't on the API free tier. Per token it's pricier than Gemini 4 Carbon until 30 November and cheaper from 1 December, when Carbon's listed price doubles.

Key facts

  • Released: 18 November 2026 (Google blog)
  • Model ID: gemini-4-neon (stable)
  • Price: $3.00 input / $15.00 output per 1M tokens; cached input $0.15 per 1M (pricing page)
  • Free tier: no
  • Context window: 1,048,576 tokens, shared by input, thinking and output
  • Max output: 65,536 tokens per request, up to 1M with long decode
  • Thinking levels: low, medium, high (default), xhigh
  • Inputs / output: text, image, video, audio, PDF in; text out

What is Gemini 4 Neon?

Neon is the new middle model in Google's Gemini 4 family. In the launch post, Google DeepMind's Koray Kavukcuoglu describes it as "bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows."

Google's What's new in Gemini 4 Neon guide places it between Gemini 4 Hydrogen and Gemini 4 Carbon. Google pitches it as frontier-level quality at a lower cost per task than Carbon, aimed squarely at coding agents and heavy knowledge work.

What are Gemini 4 Neon's specs?

From the Gemini 4 Neon model page:

  • Output is text only. It doesn't generate images or audio, and it doesn't support the Live API.
  • Long outputs: beyond 65,536 tokens, long decode continues the response in 64K-token hops, up to the 1M window.
  • Tools: function calling, structured outputs, code execution, computer use, file search, URL context, and grounding with Google Search and Google Maps are listed as supported.
  • Consumption options: Batch, Flex and Priority inference are supported.

How much does Gemini 4 Neon cost?

Google's Gemini API pricing page lists Neon at $3.00 per million input tokens and $15.00 per million output tokens, and output includes thinking tokens. Cached input is $0.15 per million, which the launch post calls a 95% discount. You need a paid-tier project.

Here's Neon next to the rest of the Gemini 4 family, from the same page (standard tier, USD per million tokens):

ModelInputOutputFree tier?
Gemini 4 Hydrogen$1.50$7.50Yes
Gemini 4 Neon$3.00$15.00No
Gemini 4 Carbon / Argon (until 30 Nov 2026)$2.00$10.00No
Gemini 4 Carbon / Argon (from 1 Dec 2026)$4.00$20.00No

What does one request cost? A worked example

To make the per-token prices concrete, take one large request with 100,000 input tokens and 10,000 output tokens (thinking included), with no caching and no grounding. Using the listed prices:

ModelInput costOutput costTotal
Hydrogen$0.15$0.075$0.225
Neon$0.30$0.15$0.45
Carbon or Argon (until 30 Nov)$0.20$0.10$0.30
Carbon or Argon (from 1 Dec)$0.40$0.20$0.60

That's my own arithmetic from Google's price list, and it assumes each model uses the same number of tokens, which they won't. Google says that at its default high thinking level, Neon has "a 32% lower cost per task than Carbon" because it gets the job done with fewer tokens. That's Google's own measurement, so test it on your workload before you trust it.

Gemini 4 Neon vs Carbon, Argon and Hydrogen: which should you use?

Google's guidance in the Neon guide and the models page breaks down like this:

  • Carbon: your hardest architecture, debugging, science and security problems.
  • Neon: complex engineering, knowledge work and science tasks where you also care about speed.
  • Argon: described as Google's efficient Gemini 4 frontier reasoning model, with faster responses and lower token spend per turn than Carbon. It's priced the same as Carbon.
  • Hydrogen: stays the default for everyday coding and agent work, and it's the only one of the four with a free tier.
  • Gemini 3.5 Flash-Lite: for high-volume, low-latency jobs.

One customer quote is useful here. Glean's AI product lead Thai Tran says that in Glean's early evaluations, Neon "matched or exceeded Gemini 4 Carbon on overall quality at 3x the speed and 60% of the cost." That's one company's internal test, but it matches how Google is positioning the model.

What thinking levels does Gemini 4 Neon support?

Neon supports four thinking levels: low, medium, high (the default) and xhigh. Google recommends starting with high for most complex work and saving xhigh for hard, multi-step problems where high falls short, since it costs more tokens and time.

This catches people who switch from Carbon or Argon. According to the migration notes, those two default to medium and cap at high. Change the model ID to Neon without setting thinking_level and your requests will think harder by default, which means more thinking tokens on the bill. Google's advice is to run your evals at all four levels and pick the lowest one that meets your quality bar.

How good is Gemini 4 Neon on benchmarks?

All of these numbers come from Google's launch post, which links each benchmark:

Google also says Neon can act as an orchestrator that directs parallel Hydrogen subagents in Google Antigravity, that it "cuts down on redundant file reads", and that it handles context compaction in long sessions. If you run long agent loops, those are the claims to check first.

Where can you use Gemini 4 Neon?

According to the announcement, Neon started rolling out on 18 November in:

  • The Gemini API (paid tier) in Google AI Studio and in Android Studio.
  • Google Antigravity, for paid Google AI Pro and Ultra subscribers.
  • The Gemini app model picker, for Pro and Ultra subscribers.
  • Gemini for business, through the API and the model drop-down.

A cyber-focused version is going to eligible defenders through Google's Fairwind Program over the following days.

How do you switch an existing app to Gemini 4 Neon?

The migration checklist is short if you're already on Gemini 4: change the model string to gemini-4-neon, check your thinking level, and make sure the project is on the paid tier. No request schema changes are needed.

Coming from Gemini 3.8 Flash or older, there's more to do:

  • Raise low output caps. The 1M window is shared, and the single-request default and cap is 65,536 tokens. Small hardcoded max_output_tokens values can cut answers short.
  • Drop explicit caching. Neon doesn't support cachedContents. Implicit caching runs automatically on prompts of at least 4,096 tokens.
  • Remove sampling parameters. temperature, top_p and top_k are ignored when in range, and out-of-range values return a 400 error.
  • Don't use minimal. Neon rejects that thinking level with an error.
  • Declare tools natively. Google warns that prompting for raw-text JSON tool calls without declared tools can return MALFORMED_FUNCTION_CALL.

Should you switch to Gemini 4 Neon?

  • Prototyping on the free tier? Stay on Hydrogen. Neon has no free tier.
  • Running Carbon or Argon in production? Test Neon this month. Their listed price doubles on 1 December, and Google claims Neon gets similar results for less per task.
  • Running long coding agents? Neon is aimed at you. Start with high, compare cost per finished task rather than per token, and only reach for xhigh when you need it.

FAQ

When was Gemini 4 Neon released?

Google announced it and began rolling it out on 18 November 2026.

How much does Gemini 4 Neon cost?

$3.00 per million input tokens and $15.00 per million output tokens on the Gemini API paid tier, with cached input at $0.15 per million.

Is Gemini 4 Neon free?

Not in the API. Google's pricing page says it isn't available on the free tier. In the Gemini app and Antigravity it's tied to Google AI Pro and Ultra plans.

Is Gemini 4 Neon better than Carbon?

Google positions Carbon as its most capable model for the hardest problems, and Neon as near-frontier quality that's faster and cheaper per task. Google reports a 32% lower cost per task than Carbon at the default thinking level.

What's the difference between Gemini 4 Neon and Argon?

Argon is priced like Carbon ($2/$10 until 30 November, $4/$20 from 1 December) and defaults to the medium thinking level, capped at high. Neon costs $3/$15, defaults to high and adds xhigh.

What is the Gemini 4 Neon model ID?

gemini-4-neon.

How big is the Gemini 4 Neon context window?

1,048,576 tokens, shared across input, thinking and output. Single responses default to a 65,536-token cap, and long decode can go further.

Sources