The token efficiency improvement might be underrated. If the model solves tasks with fewer tokens, that directly translates into lower cost and faster responses for anyone building on the API.
GPT-5.4
671–680 of 868 posts
Re: GPT-5.4
#672Earlier quoted context omitted.
Three random names isn't ideal. I'm often need to double check which is which. This is why we use numbers
They aren't random. Opus's are very long poems, haikus are very short ones (3 lines), sonnets are in between (~14 lines)
What's next? Claude Iliad?
Re: GPT-5.4
#673"GPT‑5.4 interprets screenshots of a browser interface and interacts with UI elements through coordinate-based clicking to send emails and schedule a calendar event." They show an example of 5.4 clicking around in Gmail to send an email. I still think this is the wrong interface to be interacting with the internet. Why not use Gmail APIs? No need to do any screenshot interpretation or coordinate-based clicking.
It feels like building humanoid robots so they can use tools built for human hands. Not clear if it will pay off, but if it does then you get a bunch of flexibility across any task "for free". Of course APIs and CLIs also exist, but they don't necessarily have feature parity, so more development would be needed. Maybe that's the future though since code generation is so good - use AI to build scaffolding for agent in…
I think it's akin to self driving cars prioritizing nornal roads rather than implementing new infrastructure. Tricky, but if you get it right the whole world opens up, since you don't depend on others to adapt your system.
Re: GPT-5.4
#674Results 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).
Impressive! Do you include puzzles released before the training data cutoff date?
Re: GPT-5.4
#675Earlier quoted context omitted.
It's not particularly about x.com, hundreds of site like x, youtube, facebook, linkedin, tiktok etc surreptitious add tracking parameters to their links. The iOS Messages app even hides these tracking parameters. I don't like being surreptitiously tracked online and judging by the success of my free app, there are millions of people like me.
so, since these companies have to comply with removing PII, is the worst thing that could happen to me, that I get ads that are more likely to be interesting to me? i’m not being facetious, honest question, especially considering ads are the only thing paying these people these days
The more data they have on you, the more valuable that data is to a third party. So they sell your data to someone else, who then phones you based on your known deep interest in . Or spams you. Or messages you. Or whatever method they think will most get your attention.
If you don't give them that information, they can't sell it, and the buyers won't annoy you.
It's not that the ads you get are more interesting, it's that you get more ads because they think they know more about you.
Re: GPT-5.4
#676[dead]
Re: GPT-5.4
#677In my day-to-day coding work, the top 3 coding agents are already good enough for me. On SWE-bench Verified, mini-SWE-agent + GPT-5.2 Codex is 72.8. I don’t see a comparable GPT-5.3 Codex number there, so I’m using 5.2 as the baseline. On OpenAI’s GPT-5.4 page (SWE-Bench Pro, Public), the score improves from 55.6 (GPT-5.2) to 57.7 (GPT-5.4), which is about +2.1 points. It’s a different benchmark, so this is only a ro…
I would recommend https://swe-rebench.com for comparison. It is always based on new problems.
Re: GPT-5.4
#678The question is still: Does it make your code better or worse? Only Opus makes it better, the rest worse. That's the treshold
I've been using it for three hours and it's insanely good. It's almost perfectly (needed a single touchup prompt) completed a full css refactoring that I've wanted to do for months that I've tried to have other models do but nothing worked without heavy babysitting.
Also, in the course of coding, it's actually cleaning up slop and consolidating without being naturally prompted.
Re: GPT-5.4
#679Earlier quoted context omitted.
Weird, I have had the opposite experience. Codex is good at doing precisely what I tell it to do, Opus suggests well thought out plans even if it needs to push back to do it.
This is just the stochastic nature of LLM's at play. I think all of the SOTA models are roughly equivalent, but without enough samples people end up reading into it too much.
There's a certain amount of variance in the way that people utilize these agents. Put five people in a room and ask them to compose the same prompt and you have five distinct prompts. Couple this with the fact that models respond better/worse to certain prompts depending on the stylistic composition of the prompt itself. And since people tend to write in the same style, you'd get people who have more luck with one model over another, where one model happens to align more readily with their prompt style.
To wit, I have noticed that I tend to prefer Codex's output for planning and review, but Opus for implementation; this is inverted from others at work.
Re: GPT-5.4
#680The 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…
It's a little hard to compare, because Claude needs significantly fewer tokens for the same task. A better metric is the cost per task, which ends up being pretty similar. For example on Artificial Analysis, the GPT-5.x models' cost to run the evals range from half of that of Claude Opus (at medium and high), to significantly more than the cost of Opus (at extra high reasoning). So on their cost graphs, GPT has a con…
But does it use the same agent harness? Because the harness determines the behavior a lot.