Deep-research-max-preview-04-2026 consuming millions more tokens than it should?

Google AI Studio
, , , ,
1

Hello,

Last week, I tested out the deep-research-max-preview-04-2026 agent in AI studio.

After what seemed like a long time, I discovered that the report used a total of 23M tokens, with 22M of those being input tokens. This seems quite excessive.

When I checked our spend, I found the following:

  1. Gemini 3.1 Pro: $14.76
  2. Gemini 2.5 Pro: $118.19

My question is twofold:

  1. Should the deep research max preview model be using so many tokens?
  2. Why are we being billed for 2.5 Pro when, according to the agent description, the model is supposed to be using the Gemini 3 series?

Thanks in advance.

2

Hi @Nicholas_Borg

Thank you for sharing this, can you DM me your project number ?

3

Hi @Mustan_lokhand, I don’t seem to have DM function enabled on the forum.

Is there another way I can share the project number with you?

Thanks!

4

Nicholas, I asked some questions internally about this. Although the 23M tokens is on the high end, it’s not outside of the expected bounds.

I think the 3.1 Pro / 2.5 Pro charges should be unrelated (I believe Deep Research should be it’s own billing line item).

5

Thanks for letting me know. On the 3.1 Pro Charges, I don’t get any spend data when filtering by the Deep Research agent.

6

Hi, I am also not seeing Deep Research spend and usage data in AI studio, other than total credit balance going down. The balance changes seem reasonable based on response metadata that I logged myself, but AI studio doesn’t show spend data or API logs for Deep Research requests.