Gemini cli taking large tokens or else /stats showing wrong data

[venkat9507]

What happened?

1. Created new folder to make sure i don't have any folders or files for the gemini to cache or send it as a prompt so created new folder it as nothing in it

2. Opened gemini cli running in latest version --> 0.11.3

3. Model used --> gemini-2.5-flash

4. 1st prompt given --> "Hi"

5. gemini returned --> "Hi! I'm ready for your first command."

6. i'm shocked to see --> /stats model

╭──────────────────────────────────────────────────────╮
│ │
│ Model Stats For Nerds │
│ │
│ Metric gemini-2.5-flash │
│ ────────────────────────────────────────────────── │
│ API │
│ Requests 1 │
│ Errors 0 (0.0%) │
│ Avg Latency 5.5s │
│ │
│ Tokens │
│ Total 5,364 │
│ ↳ Prompt 5,328 │
│ ↳ Thoughts 25 │
│ ↳ Output 11 │
│ │
╰──────────────────────────────────────────────────────╯

total used token for "Hi" is "5,364" and it is keep on doubling when i give 2nd prompt as "Hi" then total token is

"10,728"

NOTE: "/memory" is empty

What did you expect to happen?

reduce the token count

Client information

Client Information

Run gemini to enter the interactive CLI, then run the /about command.

> /about
# paste output here

Login information

No response

Anything else we need to know?

No response

[korade-krushna]

this is due to the system prompt which gets added before the user prompt increasing the number of overall tokens

[venkat9507]

this is due to the system prompt which gets added before the user prompt increasing the number of overall tokens

yeah understood that but not sure why it is burning that much token for every prompt it is taken 5,000 system prompts which is burning , i didn't run the cli locally and test it if your doing that can you check what kind of system prompt they are using that is taking 5k tokens ?

[scidomino]

Each request requires tokenizing:

• The built in gemini cli system prompt
• The descriptions+schema for standard gemini tools
• The GEMINI.md files you have in your CWD
• The descriptions+schema for any MCP server you have configured.
• your whole message history
• your latest message

So this is working as expected.