metir
metir
Docs
Download on App StoreGet it on Google PlayLog inSign up
Back to Blog
Gemini 4 Neon
Google
AI Pricing
Cost Per Task
LLM Benchmarks

Gemini 4 Neon at $3/$15: Price Per Task vs Carbon

Google's Gemini 4 Neon launched November 18, 2026 at $3 and $15 per million tokens. What the 32% lower cost per task claim means next to Hydrogen, Carbon and GPT-6.2.

Metir AI TeamNovember 22, 20266 min read
Gemini 4 Neon at $3/$15: Price Per Task vs Carbon

Google introduced Gemini 4 Neon on November 18, 2026, and the headline is a pricing claim as much as a capability one. Per Google's launch post, Gemini 4 Neon costs 32% less per task than Gemini 4 Carbon at its default high thinking setting, while sitting between Carbon (hardest, long-horizon work) and Hydrogen (fast everyday execution) in the lineup. This post looks at what that means for Gemini 4 Neon pricing, and why cost per task and price per token are different numbers.

$3 / $15Neon price per 1M input / output tokensGoogle's pricing page
32%Lower cost per task than Carbonat high thinking, per Google
1MToken context windowper Google
$0.15Cached input per 1M tokensa 95% discount, per Google
Google logoGoogle
OpenAI logoOpenAI
Anthropic logoAnthropic
Google's launch post refers to unnamed competing frontier models; this post compares only published list prices.

What Google published about Gemini 4 Neon

The launch post lists a 1 million token context window and pricing of $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens. Neon is available in the Gemini API through Google AI Studio (paid tier) and Android Studio, in Gemini for business, in Google Antigravity for paid AI Pro and Ultra subscribers, and in the Gemini app for paid subscribers. Eligible Fairwind Program members get access for cyber defense work.

Google reports a set of results it describes as its own or as third-party leaderboard placements:

  • Artificial Analysis Intelligence Index of 61 at xhigh, third overall, "just behind larger, slower models"
  • Vals Index of 70.9%, second, on the Pareto frontier
  • Artificial Analysis Coding Agent Index of 58.6, first overall
  • Terminal-Bench 4.1 at 65.2% and FrontierCode 2.0 at 46.2%
  • Cyber results including 90.5% on SEC-bench Pro and 51.26% on WebExploitBench v1.2, both described as state of the art

These are vendor-reported figures. Google's post also presents a comparison table as an image, so its rows are not available as text. The post does not name Anthropic or OpenAI models, referring only to competing frontier models. We therefore do not repeat any win-count against specific rivals, and any such tally should be checked against the table itself.

Gemini 4 Neon pricing next to the rest of the line

Google's Gemini API pricing page lists the four Gemini 4 models side by side, in USD per million tokens:

ModelInputOutputCached input
Gemini 4 Hydrogen$1.50$7.50$0.075
Gemini 4 Argon$2.00 through Nov 30, then $4.00$10.00 through Nov 30, then $20.00$0.10, then $0.20
Gemini 4 Carbon$2.00 through Nov 30, then $4.00$10.00 through Nov 30, then $20.00$0.10, then $0.20
Gemini 4 Neon$3.00$15.00$0.15

Two points follow from the page. First, Carbon and Argon list identical rates, and both are on promotional pricing that ends November 30, 2026. Second, from December 1, 2026 Neon's list rate is 25% below Carbon's ($3 against $4 on input, $15 against $20 on output, our arithmetic). Until then, Carbon's promotional rate is lower than Neon's per token, so Google's 32% per-task figure is not a statement about the price list alone. It also depends on how many tokens each model spends reaching an answer.

For earlier coverage of the family, see our posts on Hydrogen's launch and pricing, Argon and Carbon's cost and accuracy trade-off.

Comparing with GPT-6.2 list prices

OpenAI's model pages list GPT-6.2 Sol at $2 input, $0.10 cached input and $10 output per million tokens, and GPT-6.2 Astra at $8 input, $0.40 cached and $40 output. Both list a 1,050,000-token window, and both charge more above 272K input tokens. See our GPT-6.2 release post for the full picture.

On list price alone, Neon sits between Sol and Astra: 1.5 times Sol's rates and 37.5% of Astra's (our arithmetic). A worked example makes this concrete. A request with 50,000 input tokens and 5,000 output tokens, uncached, costs the following at list price:

One request at list price: 50K tokens in, 5K tokens out

Uncached USD cost, our arithmetic from published per-million-token rates. Gemini models in green.

Gemini 4 Hydrogen$0.113
Gemini 4 Carbon / Argon (through Nov 30)$0.150
GPT-6.2 Sol$0.150
Gemini 4 Neon$0.225
Gemini 4 Carbon / Argon (from Dec 1)$0.300
GPT-6.2 Astra$0.600

Per-request cost ignores how many thinking tokens each model spends, which is what cost per task captures.

Google logo on a glass office building at the Googleplex in Mountain View, California, seen through trees
A Google building at the Googleplex in Mountain View, California. An archive photo, not taken at the Neon launch. Photo: David Nagle, CC BY-SA 4.0.

Why cost per task is the harder number

The chart assumes equal token counts, which is exactly what differs between models. A reasoning model that thinks longer bills more output tokens for the same answer. Google's post makes three cost-per-task claims, all self-reported:

  • At high thinking, 32% lower cost per task than Carbon, with a high Intelligence Index score.
  • At high on the Vals Index and Agent Arena, roughly 50% to 80% lower cost per task than top competing frontier models.
  • At xhigh, 36% or more less per task than competing models at their highest settings.

The post does not publish the task counts, thinking budgets or rival model names behind these ranges in text. Treat them as hypotheses to test on your own workload.

“

A lower price per token and a lower price per task are different claims, and only the second one shows up on your invoice.

How to test the claim yourself

  1. Replay a real workload. Run 50 to 100 representative tasks through Neon, Carbon and Hydrogen at the same thinking setting.
  2. Log billed tokens, not list prices. Record input, cached input and output tokens per task, including thinking tokens.
  3. Price at the date you will pay. Carbon's and Argon's rates change on December 1, 2026.
  4. Score quality alongside cost. A cheaper run that fails more often costs more after retries.

Teams that route different tasks to different models, which a model-agnostic workspace such as Metir is built for, can run this comparison without rewriting their integration.

Takeaways

  • Neon lists at $3/$15 per million tokens, between Hydrogen ($1.50/$7.50) and Carbon's standard post-promotion rate ($4/$20).
  • Google's 32% lower cost per task versus Carbon is vendor-reported, at high thinking.
  • Rival comparisons in Google's post are in an image and name no models in text, so we do not repeat them.
  • Measure on your own tasks before committing.

Sources:

  • Google: Introducing Gemini 4 Neon
  • Google AI Studio: Gemini API pricing
  • OpenAI: GPT-6.2 Sol model page
  • OpenAI: GPT-6.2 Astra model page

Image credits

  • Hero: Google logo on a Googleplex building, Mountain View. Photo by Asoundd, Wikimedia Commons, CC BY-SA 4.0.
  • Body figure: Google building at the Googleplex. Photo by David Nagle, Wikimedia Commons, CC BY-SA 4.0.

Ready to experience AI that adapts to you?

metir brings together the world's best AI models in one seamless experience. Start for free today.

Get Started Free
metir

metir is where work gets done. Hire your team of Orbs, AI teammates with their own computer, alongside your mail, meetings and workspace.

© 2026 metir. All rights reserved.

Product

  • Features
  • Pricing
  • Research
  • Docs
  • Blog
  • Enterprise

Company

  • Docs
  • Support
  • Careers

Legal

  • Terms of service
  • Privacy policy

Personalisation is powerful. Privacy is non-negotiable.

Status: All systems operational