ChatGPT and API results can differ even with the same model
GPT-4o worked well in the interface for estimating the volume of items from images, but the same model through the OpenAI API gave different answers on repeated runs with the same image. The API results also differed from the and were less accurate.
The goal was to build an app that estimates item volume from . Changing API settings and trying many did not produce the desired or accuracy.
Key points
- GPT-4o performed well for image-based volume estimation in the interface.
- The OpenAI API gave inconsistent answers for the same .
- The API output was different from the output and less accurate.
- Changing API and did not solve the issue.
- Agent and app builders should validate results in the actual API setup before relying on them.