I’m using Deep Research by API with websearch tool. It counts way more Input tokens than it should be. Even with small input prompt, I get couple millions of tokens usage. In the screenshots you can see, that I have too much cached and noncached input tokens usage which I never used. And as a consequence, it burned a lot of money in my OpenAI account. It looks like bug in Deep Research api.
Customer support is ignoring me. Please help!
Welcome to the OpenAI dev community, @Rustemov_Bauyrzhan.
The deep research models are agentic models, which per the Deep Research docs:
The
o3-deep-research
ando4-mini-deep-research
models can find, analyze, and synthesize hundreds of sources to create a comprehensive report at the level of a research analyst.
Thus, the input costs would correspond to the web search result ingestion and analysis.
Yes, I know. But I’m talking about high input tokens consumption. And as you can see there are reasoning tokens usage calculation.
And also, over 2 mln input tokens usage with one small query - do you think it is ok?
I have been talking about this too .. here. O3-deep-research - 1 million tokens spent .. no output :( - #28 by jlvanhulst
have you found that the Deep Research though the API is more capable than Deep Research on chatgpt?
Mostly the same, in case of api you can manage context.
this doesn’t really answer the problem, since i’m frequently encountering situations where millions of input tokens are being consumed, but i’m getting fewer than 10 sources in the output.