Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GPT-5.4

openai.com

Post reply on HN
Print view View original

681–690 of 868 posts

Re: GPT-5.4

#681

Post by aurareturn »

peterspath wrote: post #656
minimaxir wrote: post #8

The marquee feature is obviously the 1M context window, compared to the ~200k other models support with maybe an extra cost for generations beyond >200k tokens. Per the pricing page, there is no additional cost for tokens beyond 200k: https://openai.com/api/pricing/ Also per pricing, GPT-5.4 ($2.50/M input, $15/M output) is much cheaper than Opus 4.6 ($5/M input, $25/M output) and Opus has a penalty for its beta >200…

Grok has a 2M context window for most of their models. For example their latest model `grok-4-1-fast-reasoning`: - Context window: 2M - Rate limits: 4M tokens per minute, 480 requests per minute - Pricing: $0.20/M input $0.50/M output Grok is not as good in coding as Claude for example. But for researching stuff it is incredible. While they have a model for coding now, did not try that one out yet. https://docs.x.ai/…

What kind of research do you use it for?

Re: GPT-5.4

#682

Post by aurareturn »

Alifatisk wrote: post #379

So let me get this straight, OpenAi previously had an issue with LOTS of different models snd versions being available. Then they solved this by introducing GPT-5 which was more like a router that put all these models under the hood so you only had to prompt to GPT-5, and it would route to the best suitable model. This worked great I assume and made the ui for the user comprehensible. But now, they are starting to in…

  Who’s to blame for this ridiculous path they are taking? I’m so glad I am not a Chat user, because this adds so much unnecessary cognitive load.
Most people have it on auto select I'm assuming so this is a non issue. They keep older models active likely because some people prefer certain models until they try the new one or they can't completely switch all the compute to the new models at an instance.

Re: GPT-5.4

#686

Post by crorella »

softwaredoug wrote: post #127
yanis_t wrote: post #69

These releases are lacking something. Yes, they optimised for benchmarks, but it’s just not all that impressive anymore. It is time for a product, not for a marginally improved model.

The products are the harnesses, and IMO that’s where the innovation happens. We’ve gotten better at helping get good, verifiable work from dumb LLMs

underrated comment, this is going to be the main differentiator going forward, the more powerful and versatile harness the more the models will be able to achieve and better/more advanced products will come out of it.

Re: GPT-5.4

#687

Post by salomonk_mur »

applfanboysbgon wrote: post #644
Alifatisk wrote: post #379

So let me get this straight, OpenAi previously had an issue with LOTS of different models snd versions being available. Then they solved this by introducing GPT-5 which was more like a router that put all these models under the hood so you only had to prompt to GPT-5, and it would route to the best suitable model. This worked great I assume and made the ui for the user comprehensible. But now, they are starting to in…

The real problem that OpenAI had was that their model naming was completely incomprehensible. 4.5, o3, 4o, 4.1 which is newer than 4.5. It was a complete clusterfuck. The blowback on that issue seems to have led them to misidentify the issue, but nobody was really asking for a single router model. Having a number of sequentially numbered and clearly labelled models is not actually a problem.

Having both o4 and 4o. Really. What the fuck?

Re: GPT-5.4

#688

Post by jstummbillig »

Alifatisk wrote: post #379

So let me get this straight, OpenAi previously had an issue with LOTS of different models snd versions being available. Then they solved this by introducing GPT-5 which was more like a router that put all these models under the hood so you only had to prompt to GPT-5, and it would route to the best suitable model. This worked great I assume and made the ui for the user comprehensible. But now, they are starting to in…

> Who’s to blame for this ridiculous path they are taking?

Variability, different pressures and fast progress. What's your concrete idea for how to solve this, without the power of hindsight?

For example, with the codex model: Say you realize at some point in the past that this could be a thing, a model specifically post-trained for coding, which makes coding better, but not other things. What are they supposed to do? Not release it, to satisfy a cleaner naming scheme?

And if then, at a later point, they realize they don't need that distinction anymore, that the technique that went into the separate coding model somehow are obsolete. What option do you have other than dropping the name again?

As someone else pointed out, the previous problems were around very silly naming pattern. This sems about as descriptive as you can get, given what you have.

Re: GPT-5.4

#689

Post by kinderjaje »

zone411 wrote: post #359

Results from my Extended NYT Connections benchmark: GPT-5.4 extra high scores 94.0 (GPT-5.2 extra high scored 88.6). GPT-5.4 medium scores 92.0 (GPT-5.2 medium scored 71.4). GPT-5.4 no reasoning scores 32.8 (GPT-5.2 no reasoning scored 28.1).

I added that info on https://automatio.ai/models/gpt-5-4

Re: GPT-5.4

#690

Post by rezonant »

smusamashah wrote: post #372
minimaxir wrote: post #8

The marquee feature is obviously the 1M context window, compared to the ~200k other models support with maybe an extra cost for generations beyond >200k tokens. Per the pricing page, there is no additional cost for tokens beyond 200k: https://openai.com/api/pricing/ Also per pricing, GPT-5.4 ($2.50/M input, $15/M output) is much cheaper than Opus 4.6 ($5/M input, $25/M output) and Opus has a penalty for its beta >200…

Gemini already has 1M or 2M context window right?

Yes, 1M context window since Gemini 1.5 Pro first previewed in February 2024.
Post reply on HN