Cost forecast for usage of GPT 4 endpoints about gpt-engineer HOT 4 CLOSED

It's hard to give a meaningful estimate in advance of how many output tokens will be used, although the number of input tokens would be known.

Asking if they want to continue would also need another ask for confirmation step, which I personally would always want to skip, thus would need some flag or config to disable, adding further complexity.

I question if most people really need/want this?

If users are on a limited budget, they could just set their spend limit in the OpenAI API settings.

from gpt-engineer.

It's hard to give a meaningful estimate in advance of how many output tokens will be used, although the number of input tokens would be known.

Asking if they want to continue would also need another ask for confirmation step, which I personally would always want to skip, thus would need some flag or config to disable, adding further complexity.

I question if most people really need/want this?

If users are on a limited budget, they could just set their spend limit in the OpenAI API settings.

I'm completely with you on not having unnecessary prompts for the user. Here's why i thought of my suggestion and thought it might be good because no one likes cost surprises.

Yesterday, I was creating an informational video on this tool and during the process i executed it to generate some simple codebase. And it did perfectly, but to find API cost : $0.05 or something at the end of the execution was a 'little surprise' to me. (I have used the tool earlier but completely forgot it uses GPT 4 API key behind the scenes.

So I thought it would be a 'nice-to-have'. Otherwise, what if i instruct it to generate a fairly complex codebase and it gives me a bill of say $5 unexpectedly? still not a lot but I hope I conveyed my point.

Setting the budget on Open AI itself is not relevant i guess given it's not even possible to track usage by API key

• https://community.openai.com/t/does-openai-support-seeing-cost-per-api-key-and-allow-limits-to-be-set-at-api-key-level/194418/2
• https://community.openai.com/t/how-to-track-api-usage-and-cost-by-api-key/174145/42).

Regarding the meaningful estimate, it can be ballpark then user has some idea than none.

I'm very much open to discussion on this.

from gpt-engineer.

Hi,
@RukshanJS thanks for the idea. From other comments it seems like budget is top of mind for many users, so I think your thoughts are relevant. For some combinations of flags, where the number of executions are known, it would be possible to provide an upper bound for the cost.

I see two problems:

1. I agree with @ErikBjare that it will complicate the workflow. One possible way would be to print the number directly and then tell the user to abort the run dynamically if the price tag is too much. This would bypass the additional confirmation step.
2. We are currently low on people developing features and the core team is focused on the most pressing functionality. If you are willing to implement this yourself, we should consider it.

from gpt-engineer.

Hi, @RukshanJS thanks for the idea. From other comments it seems like budget is top of mind for many users, so I think your thoughts are relevant. For some combinations of flags, where the number of executions are known, it would be possible to provide an upper bound for the cost.

I see two problems:

1. I agree with @ErikBjare that it will complicate the workflow. One possible way would be to print the number directly and then tell the user to abort the run dynamically if the price tag is too much. This would bypass the additional confirmation step.
2. We are currently low on people developing features and the core team is focused on the most pressing functionality. If you are willing to implement this yourself, we should consider it.

HI, sorry for the late reply. Yes I would love to work on this. I'm still learning this stuff so would love to take tasks like this. Let me know.

from gpt-engineer.