the input offset current
The two inputs of an op-amp each draw a bias current, but those two currents are never perfectly equal. The input offset current is the difference between them — the mismatch left over after you have accounted for the average. If bias current is how thirsty the two inputs are on average, offset current is how unequally thirsty they are.
Ios is defined as the magnitude of the difference between the two input bias currents: Ios = |Ib+ minus Ib-|. It is usually only 10 to 25 percent of the bias current itself. This matters because the classic trick of adding a matched resistor in each input cancels the EQUAL part of the bias current, but it cannot cancel the part that is unequal. Example: with Ib = 100 nA and Ios = 10 nA, matched 1 MΩ resistors remove the 100 mV from the average bias current but leave 10 nA times 1 MΩ = 10 mV of stubborn residual error.
Why this matters: once you have done everything right — matched your source resistances to cancel bias current — the offset current sets the true error floor of a high-impedance front end. For precision work with large resistances, you choose a part not just for low bias current but for low offset current, and you keep both input resistances equal so the mismatch is the only thing left to fight.
An instrumentation front end uses matched 1 MΩ resistors so the 50 nA bias current cancels. With an offset current of 5 nA, an irreducible 5 nA times 1 MΩ = 5 mV error remains — and that, not the bias current, is what you must trim or calibrate.
After matching resistors, offset current — not bias current — is the residual error.
FET-input op-amps have such tiny bias currents that the matching resistor is often pointless — its own thermal noise and the bias current's mismatch can do more harm than good. Match resistors mainly with bipolar-input parts.