I'm actually not sure what will become of tools like DeepL. Whatever edge they may have with dataset tuning and other tricks under the hood are likely supersede

I'm actually not sure what will become of tools like DeepL. Whatever edge they may have with dataset tuning and other tricks under the hood are likely superseded by a better architecture, which in turn requires a ton of capital to train. By the time they come up with a GPT4 equivalent, we will be using GPT5.