is there a cheaper way to run GPTCelltype?

[jasonmadore]

I am currently running this across a number of projects because it can save a lot of time but the API costs are becoming prohibitive. I spent 100$ in 1 day running GPTCelltype to label the results from a loop of seurat clustering... oops. Any suggestions for running this cheaper?

[jasonmadore]

The cheaper alternative obviously is to use a cheaper gpt model.

res <- gptcelltype(marker_genes, model = 'gpt-4o-mini', topgenenumber = 10, tissuename = 'human lung cancer')

1 million cells across 134 subclusters cost 'gpt-4o-mini' only 0.07$ whereas gpt-4.0 cost 15.00$.

[parashardhapola]

With CyteType you don't have to pay, you can run 3 annotations per day for free. What's even better, they have heavily benchmarked 16 LLMs to compare the performance of cheaper and more expensive models.

https://github.com/NygenAnalytics/CyteType