Thank you for all the feedback and reports on this issue.
We are working on addressing this issue at the model-level, and plan to share an update in January.
In the meantime, we will prioritize stability of 1106 models and will not be changing its behavior. This means that 1106 models will continue to produce unicode escape sequences (i.e. \u006c). We believe that the standard libraries of most languages will be able to handle this format when parsing the API output to JSON. See our docs for details.
If this is a critical issue for your use-case, we recommend that you use 0613 models in production for now.