How likely is it for a small company to develop a model that outperforms the big ones (GPT, Bard etc)?

There are 3 players in the AI space right now. All purpose LLM titans (Google, OpenAI, Meta), fancy domain specific apps that consume one of the big LLMs under the hood, and custom developed models.

I know how to judge the second type as they basically can do everything the first one can but have a pretty GUI to boot. But what about the third ones? How likely is it for a (www.yet-another-ai-startup.ai) sort of company to develop a model …

apps artificial bard big etc google gpt judge llm llms meta openai small small company space