Skip to content
Developer312

Gemini 4 Neon's $3 and $15 Rates Are Not a Task Price

Google's November 18 post lists Gemini 4 Neon at $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window. The lower cost-per-task figures in that post are Google's own comparisons.

Published Report an error

Google's November 18 post lists a price and a context window for Gemini 4 Neon in one sentence. The announcement says the model is priced at $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window. The same sentence lists cached input at $0.15 per million tokens, which Google calls a 95% discount.

Those are list prices and a capacity limit, not a bill for a finished job. Google also says that at its default high thinking level, Neon pairs a high Artificial Analysis Intelligence Index score with a 32% lower cost per task than Carbon. That comparison is Google's. The post does not show the task mix, the token counts, or the invoice a buyer would receive.

Google says Neon is rolling out in the Gemini API in Google AI Studio, in Gemini for business, and for paid Google AI Pro and Ultra subscribers in Google Antigravity and the Gemini app.

Key Takeaways: What Matters Most?

  • Google's November 18 announcement lists Gemini 4 Neon at $3 per million input tokens, $0.15 per million cached input tokens, and $15 per million output tokens.
  • Google describes the cached rate as a 95% discount and says the model has a 1 million-token context window. That window is a capacity limit, not a bill.
  • Google says that at the default high thinking level, Neon has a 32% lower cost per task than Carbon. That comparison is Google's, and it is tied to that thinking level.
  • Google separately says Neon scores 61 on the Artificial Analysis Intelligence Index at the xhigh thinking level, ranking third overall. That score is not the figure in the 32% sentence.
  • A list price is concrete enough to test. A vendor cost-per-task percentage is not a substitute for the test.

Takeaways like these, every weekday morning. Get the briefing

Token Rates Still Need a Task

Input, cached input, and output have different published rates, so the shape of the work matters. A job that rereads a long prompt is not the same purchase as a job that writes a long answer. The $0.15 cached rate applies only to tokens that actually hit the cache Google is pricing. The announcement states the discount. It does not say what share of a given application's tokens will qualify.

The 1 million-token window, on Google's account, is how much context the model can accept. It does not show that the model will use the relevant part of that context, or that a long prompt is cheaper than a shorter one that has to be retried. Cost per million tokens and cost per accepted task are different questions. The same check is the one we applied to Gemini 4 Hydrogen's published rates: rerun a fixed evaluation, and do not copy a vendor percentage into a budget.

The 32% Figure Is Google's Comparison

The 32% comparison is Google's, and it is tied to the default high thinking level and to Carbon. It is not a result for every competing model or every thinking level. Google says Neon has four thinking levels. Changing the level changes the setting, even when the model name stays the same.

Google gives a different index figure at a different level. It says that at xhigh, Neon scores 61 on the Artificial Analysis Intelligence Index, ranks third overall, and shows its highest intelligence score. That 61 is not the number in the 32% sentence. Using them as one result would collapse two thinking levels.

Google makes further cost claims in the same post, and they are also Google's comparisons. It says that at the high level, Neon sits on the Pareto frontier on the Vals index and on Agent Arena, with roughly 50% to 80% lower cost per task than top competing frontier models, and that xhigh still costs 36% or more less per task than competing models at their highest setting. None of those figures is an independent audit or your invoice. The 32% claim is a hypothesis to test on one accepted task, counting tokens, retries, and review.

What Builders Should Take From It

  • Write down the rate card. Google lists input at $3 per million tokens, cached input at $0.15, and output at $15. Keep cached and uncached traffic separate until you measure the hit rate.
  • Do not book the 32%. That lower cost per task versus Carbon is Google's comparison at the default high thinking level. Put it beside your result, not in place of it.
  • Keep the thinking level in the log. Google's score of 61 is at xhigh. The 32% comparison is at default high. Changing both at once is two changes.
  • Treat the context window as capacity. One million tokens is the stated limit. Score whether the answer was acceptable, not how much of the window you filled.

Neon's list price is concrete enough to test. Google's cost-per-task percentages are not a substitute for that test.

Developer312 covers the AI business signals builders actually need to act on. Get the weekday briefing at developer312.com.

Sources

  1. [1]Introducing Gemini 4 Neon (opens in a new tab) — Google (2026-11-18)

Frequently Asked Questions

Get the next briefing

The lead story each weekday morning, with the key takeaways and a link to the full analysis.

One lead story each weekday morning: what happened, the key takeaways, and a link to the full analysis.

New posts daily, briefing weekday mornings. We use your email only to send the briefing. Unsubscribe anytime. Privacy policy.

Related Articles