Different standard errors in reghdfe 3.2.7 versus reghdfe 5.7.3

[liaoguanmin]

Hello Sergio,
I am running a specification with reghdfe. For the same regression in which standard errors are clustered at both region and year levels. However, reghdfe 3.2.7 and reghdfe 5.7.3 give different standard errors. If I suppress the constant term in reghdfe 5.7.3, the difference becomes smaller but it still exists. Could you please tell me the reason for this? Thank you very much!
Best regards,
Guanmin