Hey everyone! I’m a customer of google’s 2.0 flash, and thought to try the experimental thinking approach. Was blown away by the performance on the AI Studio, especially with its vision capabilities! However, when giving it a shot on the API, I was pretty dissapointed. Prompts which worked with images on AI Studio would not work through the API, regardless of what temperature or top P values I used. They would sometimes work, but it felt the model just wasnt as smart as the AI studio one for vision tasks primairly. Anyone have some thoughts or suggestions? I’m using google’s native types.Part.from_bytes to send the image over in the mime type of JPEG.
Any help would be appreciated, thank you!