Date: April 08 Campany: NA Contact: Eren şahin <sahineren.09@gmail.com> Hi Nixtla Team! I finally created a spare time for testing TimeGPT. I used 78 time-series (which is a very small sample of what I have been working on, I usually train on millions of series), having 3 years of data on a weekly basis. Which makes approximately ~156 data points for each. I also used the historical forecasting option so I could evaluate the performance of TimeGPT over a long period of time where I could observe multiple seasonal patterns. Once I checked the billing section, I was surprised with the bill of the consequences of forecasting only 78 time-series. Please see the attached image below. Hope this feedback is helpful for finding your optimal pricing strategy, I find it quite expensive in its current state (1$ for each time-series). Best Regards, Eren

What would you answer?

Thanks for the interest and feedback. We are currently working on the pricing model to better suit all use cases. Don't worry about the current billing dashboard. While we figure out the real pricing, we have given you 5K in additional credits so you can keep testing. For our records, it would also be great to understand your use case. Best Diana

Great to be asking for that feedback! I imagine this person will soon get the 'please take this TimeGPT survey' email. But after we send those emails out, and if we get questions like this from new people, we could send them a link to the relevant survey to get their feedback.

Lets send that email ASAP so the users does not churn.