Hydrogen can be much cheaper for (say) the last 10%, because (1) hydrogen has very low capital cost per unit of storage capacity, and (2) the efficiency hit of going through hydrogen vs. batteries is less important when it's just 10% of the total.

Think of batteries and hydrogen as analogous to cache memory and main memory in a computer. They have different performance and economic characteristics and compensate for each others weak points.

To see this in operation, go to https://model.energy/ and try turning hydrogen off and on in the settings. If you simulate for Germany, for example, turning off hydrogen can double the cost of achieving a certain level of constant grid power. Hydrogen can be particularly valuable for places with large seasonal variation or lots of wind (which has a long timescale component in how it varies.)

I will add that China is already selling electrolysers for < $300/kW, less than half that simulation's 2030 cost assumption.