I've been looking at the pricing on OpenAI's site and noticed that o3 is not only cheaper than o1 but also seems to perform better according to various benchmarks. Given that o3 is available and has these advantages, is there still any reason to use GPT o1?
5 Answers
From what I've seen, o3 outperforms o1 in most categories, except for hallucinations. So if you're worried about that, o1 might still be worth considering. It was known to be more reliable in that aspect.
In my opinion, the hallucination issue in o3 really overshadows any supposed improvements over o1. I'm not sold on the idea that o3 or even o4 are truly 'better' in practical terms. Sure, they have new capabilities like image reasoning, but I wish they’d focus more on accuracy instead of flashy metrics. Plus, they have a different vibe compared to o1, which I found much better for straightforward interactions.
Metrics alone don’t tell the full story, especially for real-world applications. There’s still a debate out there about whether o3 is truly superior to o1, despite what the marketing suggests.
I'd say o3 is great for its search capabilities and external knowledge integration. But o1 might have a slight edge in some knowledge areas because it's a bigger model. If I need a heavier model, I usually go for 4.5 instead. Just a heads up, I've noticed o3 sometimes throws in random Chinese characters in its outputs, which can be annoying if you're looking for clean results.
Honestly, I’ve struggled to get o3 to do anything useful. Have you had better luck? Most people I talk to share the same sentiment.
What kind of tasks have you tried? I see mixed reviews, but my experience has been mostly positive with o3. It had some initial flaws, especially with Canvas, but overall, it feels stronger than o1.
I had my preferences for o1 too, but since it's gone from ChatGPT, I guess I'm making the switch to o3. Just felt like o1 gave me higher quality answers, though.