The -turbo models in the past have been much worse too. gpt-3.5-turbo is way way worse than text-davinci-003 (gpt-3.5).

The -turbos are correspondingly priced. gpt-4-turbo is ~1/3 price of gpt-4, 6.6x more expensive than gpt-3.5-turbo-instruct and 20x gpt-3.5-turbo.