That is a fair-sounding but actually invalid argument for multiple reasons:
1. OpenAI fully controls the user cost for a model, and can set it to where it sits well on the curve.
2. Performance of shrunken models like Sol/Terra/Luna is derived from the level of shrinking (relative to Astra). As such, the size and performance of the model is something that is actively targeted when developing the model. If the performance target for Terra was inappropriate for v5.6, this is no way means that it had to be this way for v6.
That is a fair-sounding but actually invalid argument for multiple reasons:
1. OpenAI fully controls the user cost for a model, and can set it to where it sits well on the curve.
2. Performance of shrunken models like Sol/Terra/Luna is derived from the level of shrinking (relative to Astra). As such, the size and performance of the model is something that is actively targeted when developing the model. If the performance target for Terra was inappropriate for v5.6, this is no way means that it had to be this way for v6.