Apply the corresponding rate to uncached input tokens, input tokens served from the implicit cache, and output tokens including thinking tokens. For the same token usage, Batch and Flex each cost half as much as Standard, while Priority costs 1.8 times as much as Standard.
Suitable tasks. Budgeting for a single Gemini Developer API request or a fixed batch workload, and comparing model costs across the four inference service modes.
Unsuitable tasks. Calculating rates for Google Cloud enterprise projects or reconciling every charge on an official bill.
Applicable model version. gemini-4-neon.
Applicable clients, agents, or APIs. The paid Gemini Developer API. Neon is unavailable on the API's free tier.
Recommended thinking levels and settings. Use Standard as the budget baseline. When comparing thinking levels, sample output usage at each level in the same service mode and include thinking tokens.
All three rates are in US dollars per million tokens.
Uncached input tokens = total input tokens − cached input tokens
Model cost (USD) = (uncached input tokens × input rate
+ cached input tokens × implicit caching rate
+ billable output tokens × output rate) ÷ 1,000,000Billable output includes thinking tokens. If your usage record already includes them in the output count, use that count directly. If the record lists thinking tokens separately from other output tokens, add the two counts first. If you have no cache usage data yet, use 0 cached input tokens in the estimate.
Suppose a request uses 100,000 input tokens, of which 60,000 hit the implicit cache, along with 5,000 answer tokens and 5,000 thinking tokens. The estimated model token cost in Standard is (40,000 × 3 + 60,000 × 0.15 + 10,000 × 15) ÷ 1,000,000 = $0.279. The same usage costs $0.1395 in either Batch or Flex and $0.5022 in Priority.
Choose the rates for the service modes available to your workload. If you need to compare thinking levels, run the same set of representative inputs at each level and record usage separately.
For each request, record total input tokens, input tokens that hit the implicit cache, and billable output tokens. Use the formula to calculate the model cost. If you use multiple service modes, calculate costs at each mode's rates and add them together.
Check tool fees separately when using Google Search or Google Maps grounding. Google Search measures usage by the number of queries it actually runs. A single Gemini request can trigger multiple searches.
| Inference service mode | Input | Output (including thinking) | Input served from the implicit cache |
|---|---|---|---|
| Standard | $3.00 | $15.00 | $0.15 |
| Batch | $1.50 | $7.50 | $0.075 |
| Flex | $1.50 | $7.50 | $0.075 |
| Priority | $5.40 | $27.00 | $0.27 |
The pricing tables for all four modes list implicit caching as the only supported caching option. Google also says that when grounding is enabled, the internal tool system instructions and function declarations the model processes count toward token charges.
The formula estimates model token costs only. The example uses hypothetical usage. The actual charge depends on the service mode, billable usage, cache hits, and tool usage. Refer to the project bill for the final amount.
On 2026-11-19, the official pricing page was opened in the international edition of Tabbit to check Neon's Standard, Batch, Flex, and Priority tabs. You can reproduce the example using the table above. The check involved no paid API calls or access to project billing records.
Gemini 4 Neon