Dobrý večer, moc se omlouvám, neumím anglicky a moc opravdu moc mě zajímá, jaký průběh a veškeré možné informace o GPT-4.5, video se mi nedalo nastavit do českého jazyka. Děkuji moc za informace, které si pak eventuálně dám do překladače. Dana
1 Like
you try it first, I’ll follow 
1 Like
I am understanding they made the model smaller somehow like the 08 being a quant or something so it is being more efficient for them for running.
1 Like
but wouldn’t that be a “mini”?
Hi!
Můžete najít blogový příspěvek a systémovou kartu na webové stránce:
2 Likes
wow, it feels very fast. 
3 Likes
I am not knowing who this model is being for. It is being a great model of course but is it being 15x better than 4o? Perhaps there are being some people with more moneys than time who not wanting to work with the model back and forth, but even having 4o and o3mini doing back and forth revisions is maybe going to be being better and much cheaper than 4.5?
It is also being not multimodal so no vision or image creation.
Is this announcement being a dud at this pricing?
Is it being 0.15$/1,000 tokens fast though?
I am thinking this will be being a big upsell for Pro with unlimited 4.5 they are just using as default with the rest of us getting a few messages each day for the 4.5.
We are still waiting for 4o image ability too 
With the pricing being on the top so we don’t even needing to read it! 
4 Likes
So it costs more than four times as much as the o1 model for input tokens, and 2.5 times as much for output tokens, huh?
I still don’t have access to the full o1 model myself, though.
As for GPT-4.5, I hope they’ll revise the pricing going forward and make it gradually more affordable.
Just checking in with those who know—is the listed pricing for this model a typo? I feel like the decimal is in the wrong place.
Anybody know for sure?
Well … slightly disappointed in one respect:
My chatbot has become more straightforward and boring and acts less in the character that I requested in the System Prompt. That’s really unfortunate, because it’s nice to give the Chatbot some character.
Let me know if you see that too …
1 Like
I slowed GPT-4.5 API to a crawl, reminiscent of GPT-4 of March 2023. And it reasoned its way through a puzzle for 1580 tokens of output = $0.24 with persistence, yet did not come up with the obscure foundation required, at a point in time within its knowledge cutoff, that so far only O1_Pro has discovered:
4.5 answered that Paul (Paul Dirac) might be the list completion needed, due to a four-letter Nobel Prize winner connection in other names, despite not being able to resolve the mismatch of Mira against that criteria.
4 Likes
Thank you for correcting me!
The APi doc for vison wasn’t yet updated to list 4.5 and I just searched the system card for vision and found nothing. There is being only 2 mentions of images in the card. One is about overrefusals and the other is just in an example problem from the MLE-Bench.
It does not seem like they are sell this model as multimodel even if it is multimodal. Maybe the vision benchmarks for 4.5 are not being what they hope?
1 Like
OK I had to move the “character defining” instructions to the bottom of the System Prompt … then the chatbot returned to acting more as I expected … curious!
1 Like
You’ve got to come up with a pretty good case for vision needing this specific model when a 1800x1000 image costs more to send to GPT-4.5 than to generate with DALL-E-3.
1 Like
I have a setting in my chatbot that allows you to switch the vision function to a separate model.
1 Like
Interesting.
New model = new prompting needed. Time to experiment and learn the model! (If I were being rich enough too 
)
I am curious how it really does compare to 4o with a great system prompt and a review pass or two?
Sorry, I am being really hung up on the 150$ million token price. That is being 10x the price of Claude 3.7 Sonnet! If Anthropic released a 3.7 Opus at the same cost of 3 Opus it would be being half as expensive!
Even if 4.5 is being the best state of the art non-reasoning model maybe I am not being smart enough to see where it fits in the world?
I agree. I do not know who will be using this model unless it is needing to be better output in only one pass.
Maybe this is being the advanced voice model in the future or maybe they can make a 4.5o-mini in 6 months?
I am just being disapointed though because I was hoping for a big move forward and this is not being that.
1 Like
If the lower price is correct … then if this can replace gpt-4-turbo (my favourite model of all of them) that’s a BIG saving (less than 33%)
2 Likes