Honestly I think it would have to:

1) Benchmark meaningfully higher than other models

2) Be offered by a cloud provider (like Azure+OpenAI / AWS+Anthropic). Otherwise you have very little track record in model/api stability. Especially looking at the last week.