Neon Field GuideIndependent reference

How is Gemini 4 Neon priced?

Budget for input, cached input and output separately. Check the live Google rate card for your billing route.

Build a request estimate

Count uncached input, cache reads and output separately. Multiply each count by its current rate per million tokens and divide by one million. Estimate retries as separate requests and include any cache storage or tool fees that apply.

Chat subscriptions and API billing are separate. Set a small test budget and use actual billed usage to refine it. Google API pricing

Calculate from your rate card

Gemini 4 Neon API rates per million tokens

These are the standard paid Gemini API rates in USD. Google lists no free API tier for Neon in this table. Output pricing includes thinking tokens, and the caching rate applies to implicit cache hits. Google lists explicit caching as unsupported for Neon.

UsageUSD per million tokensSource checked November 20, 2026
Uncached input$3.00Google Gemini API pricing
Cached input$0.15Google Gemini API pricing
Output including thinking$15.00Google Gemini API pricing

Three worked cost examples

These are estimates for illustrative token counts. They assume the standard paid rate above, with no tax, separately priced tool usage or provider markup. The input counts in each example are split into mutually exclusive cached and uncached buckets.

ScenarioUncached inputCached inputBilled outputEstimated total
Long document pass500,000050,000$2.25 USD
Repeated document pass100,000400,00050,000$1.11 USD
Smaller code review12,00003,000$0.08 USD

Long document pass. Half a million uncached input tokens cost $1.50. Fifty thousand output tokens cost $0.75. That produces a $2.25 estimate. This example is useful for a document review with a long source pack and substantial output. If the same review can be completed with a shorter response, output costs fall in proportion to the billed output tokens.

Repeated document pass. Start with the same total input and output. Assume 400,000 input tokens are reported as cached and 100,000 are uncached. The input amounts cost $0.06 and $0.30 respectively. The same output costs $0.75, bringing the total to $1.11. The saving against the uncached scenario is $1.14. The cache discount applies to the hit portion of input, so savings on the whole request depend on the output share.

Smaller code review. A request with 12,000 input and 3,000 billed output tokens costs $0.081 before rounding. One thousand requests with that exact usage would cost $81. If a share needs another attempt, add the observed billable usage for those attempts to the budget. Estimate the average from a representative set instead of assuming that every review has the same token counts.

How to read usage before estimating a batch

Open the usage details for a successful sample request. Identify total prompt tokens, cache hit tokens and billed output. If your API response includes cache hits inside the total prompt count, subtract the hits first to get uncached input. Billing output can include thinking tokens even when the visible answer is short. Use the usage fields documented for the particular endpoint because different clients may present the same totals under different labels.

Run the sample on files that resemble your normal workload. A few short examples can miss large attachments, generated tables or responses that require extended reasoning. Compare the typical request with a large one and record both. For a recurring task, note whether repeated calls reuse a long prefix and whether the service reports actual cache hits. This gives you a range for a budget and a way to check it against the billing console.

Which costs belong outside the token estimate?

Check whether your workflow uses separately priced grounding or other tools. The Google pricing page lists tool terms separately from the model's rates. If another product manages the calls, check its own bill as well. The same model can appear in a chat subscription, a developer API project or a business product with different access and billing. Use the rate card for the product and project that will receive the request.

Budget retries from usage records. An attempt that failed before processing may have a different billing outcome from a completed response that your application rejected. Failed validation, a timeout and a second prompt can create a different token profile from the original request. Track available usage per attempt and compare the sum with your bill rather than presuming every error is free or always billed in full.

What should a production budget include?

Separate the model cost from the cost of getting a result you can use. For a review queue, divide all model charges by the number of accepted reviews and track human correction time separately. A smaller request may be cheaper per call and still require more attempts. For a batch job, reserve room for your largest source packs and set budget alerts for the actual billing project.

Before committing spend, recheck Google’s standard rate table for Neon and compare the estimate with a small paid trial. Then use the Neon cost calculator to try a different cache hit share, output length or rate.