Microsoft Foundry Token Consumption

GNHD_TKR 0

Currently, on Microsoft Foundry,

the LLM is ChatGPT 4.1 Mini,

and the tools use Azure AI Search.

While testing in the playground,

the first question consumed 15,000 to 20,000 tokens. This is within expectations.

However, the second question consumed 100,000 to 300,000 tokens.

I believe until about a week ago, the second and third questions also consumed the normal amount of tokens. Did I change some setting?

Also, if deployed and used on Copilot or Teams, would it consume a similar number of tokens?

GNHD_TKR 0 Reputation points

2026-02-19T06:29:46.1633333+00:00

The above information did not resolve the issue.

I would prefer assistance from the support team, not AI.
Manas Mohanty 14,915 Reputation points Microsoft External Staff Moderator

2026-02-19T13:32:53.6533333+00:00

Hi GNHD_TKR

Total Token consumption depends on no token searched from search index and prompt token and processed token.

You can reduce token counts by explicitly mentioning in the System instruction to keep the word limit to 200 words or adjusting "Max Completion tokens" from playgrounds/parameters

Please share screenshot of token usage along the prompts in private message.

Thank you.

Your answer