Everything is same but first image is simulation done in ngspice-43 and the second one is done in ngspice-42. Ngspice-43 is causing lot of issues in simulations. My friend tried shifting to previous versions in docker but same issue.

It is a convergence problem. The bad sim result shows a -8.2 MV. This happens because you're using a ideal current source and the circuit doesn't converge naturally. Try biasing with 1 uA, for example, and see if it converges. You should increase your transistor widths, as they are too short.

Is there someother way to solve this, bcs it works in previous versions. I designed it from gmid scripts

Make a NMOS current mirror with two outputs at 10 uA and bias the PMOS with them, just to see what happens.

It works thank you