Batch API Gemini 2.5 Flash charges are abnormally excessive in AI Studio

Payment Account Number: 01B180-C4A819-FE4D1A

[Before, Aug~Oct]

Type: Batch API

Model: Gemini 2.5 Flash(Text)

Requests: 200,000~

(SKU) BatchGenerate content output token = 635,099,452 count

Billing: 758 $ ~

TIER: 1

[Current, Nov]

Type: Batch API → SAME WITH ABOVE

Model: Gemini 2.5 Flash(Text) → SAME WITH ABOVE

Requests: ONLY 2,000 → SAME INPUT PROMPT WITH ABOVE

(SKU) BatchGenerate content output token = 597,735,124 count

Billing: 750 $ ~

TIER: 3

The only changes are the TIER and the number of requests. Everything else is completely identical.
Compared to the previous setup, we requested only 1% of the Batch API calls, yet the billing came out similarly high.
This seems to be due to the BatchGenerate content output token being abnormally high.
Since we entered the same prompt, we cannot assume we made requests with long outputs, and no one else is using the API key.
I have registered a billing support case. Please explain why this result occurred.