GPT 4o vs GPT 4 – Massive improvements

OpenAI just released GPT 4o for the public and I went on and tested it right away and the results are massive! So, I just reposted some of my recent GPT 4 prompts to GPT 4o and here are the results!

First off the performance improvement is fantastic. Before GPT4’s performance was inconsistent, sluggish, and slow at best. On the left, these are the numbers from several sources. On the right the estimation from OpenAi.

GPT-4 TurboGPT-3.5 TurboGPT-4o (new model)
~50 tokens/s
~150 tokens/s
~100 tokens/s
(OpenAI’s estimate)

While the 100 tokens/s should be the 2x OpenAI promised, it feels much faster and consistently faster as well. I wouldn’t be surprised if it turns out that it performed as fast as GPT-3.5. Instinctively it does!

Example prompts

Here are a few of my recent GPT 4 queries that I ran again with GPT 4o and at first sight the responses are way better. The numbers shown by OpenAI weren’t really that impressive to be honest, but it’s quite noticeable that 4o is more capable than any previous model.

Example #1 with GPT 4 Turbo (old)

Example #1 with GPT 4o (new)

Example #2 with GPT 4 Turbo (old)

Example #2 with GPT 4o (new)

Example #3 with GPT 4 Turbo (old)

[excerpt from Wikipedia]

Example #3 with GPT 4o (new)

[same excerpt from Wikipedia]


Leave a Reply

Your email address will not be published. Required fields are marked *