I tried it with a more complex screenshot, and o4-mini-high failed, missing a few important elements and making a few dumb mistakes.It’s probably just that OpenAI models are just good at vision tasks; and somehow the FTL store screen is out of the reach of all non-OpenAI models, but most OpenAI models can read it.
I tried it with a more complex screenshot, and o4-mini-high failed, missing a few important elements and making a few dumb mistakes.
It’s probably just that OpenAI models are just good at vision tasks; and somehow the FTL store screen is out of the reach of all non-OpenAI models, but most OpenAI models can read it.