Does GPT-o3 API have think-with-image feature?

API
1

I tried GPT-o3 API for some VQA tasks, the number reasoning tokens is always less than 1k. And the performance is much worse than in the chatgpt webpage.

Did it really think-with-image as it did in the website CHATGPT page?

The image shows a split comparison of a robotic setup with objects highlighted in different colors, displaying a target area for pouring. (Captioned by AI)
The image shows a split comparison of a robotic setup with objects highlighted in different colors, displaying a target area for pouring. (Captioned by AI)1114×814 52.2 KB