(I took the liberty to reverse your numbering to the standard.)
You use the model that fits your purpose (there are others as well).
The TCP/IP model is centered on the middle layers, well suited for software developers using the IP stack. It's a "real-world" model and fits almost always. A developer doesn't care much for the physical details, so they're all mashed up under "network access".
Network engineers require a better insight into the lower layers, so they might use the OSI model for the physical and the data link layer. Usually, they don't care much about what goes on in the application layer - that's why OSI layers 5+ are off-topic on NE.
A hardware engineer requires an even more detailed view into the lower layers, so they're split into several sublayers each. IEEE Ethernet splits the physical layer in Physical Coding Sublayer, Physical Medium Attachment, and Physical Medium Dependent. The data link layer is split into Media Access Control and Logical Link Control.
Layer models are about structuring the total workload into smaller pieces that are easier to handle and better to grasp. When designing a new application, the upper OSI layers can be helpful for structuring the data and control flows, even though OSI is often said to be a theoretical, non-practical model. Last but not least, a good layering structure is key for modular development and reusable technology.