There are models that are open source and near GPT3. The most open are eleutherai's models, though not as big as GPT3 perform very well. You can go run them right now with some very basic python.

The problem is less that we don't have open models, it's that we haven't found good ways to run the models that big on consumer hardware. We do have open models that are about as big as GPT3 (The largest Bloom model) but the minimum requirements in GPUs would set you back about 100,000 us dollars.

Stable Diffusion didn't just democratize image gen AI by releasing SD open source, but by releasing it in a way people with normal gaming computers could use it.

We are maybe almost at this point with language models. Flexgen just came out, and if those improvements continue we might get an SD like moment. But until then it doesn't matter if GPT3 is open or not for the vast majority of people.