On the other hand, OpenAI has moved to a naming convention where they seem to use a name for the model: “GPT-4”, “GPT-4 Turbo”, “GPT-4o”, “GPT-4o mini”. Separately, they use date strings to represent the specific release of that named model. Whereas Anthropic had a name: “Claude Sonnet”, and what appeared to be an incrementing version number: “3”, then “3.5”, which set the expectation that this is how they were going to represent the specific versions.

Now, Anthropic is jamming two version strings on the same product, and I consider that a bad choice. It doesn’t mean I think OpenAI’s approach is great either, but I think there are nuances that say they’re not doing exactly the same thing. I think they’re both confusing, but Anthropic had a better naming scheme, and now it is worse for no reason.