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

OpenAI dropped the price of o3 by 80%

twitter.com

Post reply on HN
Print view View original

381–390 of 518 posts

Re: OpenAI dropped the price of o3 by 80%

#381

Post by tauntz »

tedsanders wrote: post #328
energy123 wrote: post #302

Earlier quoted context omitted.

Is it o3 (low), o3 (medium) or o3 (high)? Different model names have crept into the various benchmarks over the last few months.

o3 is a model, and reasoning effort (high/medium/low) is a parameter that goes into the model. o3 pro is a different thing - it’s not just o3 with maximum remaining effort.

Why's it called o3 then if it's a different thing? There's already a rather extreme amount of confusion with the model names and it's not clear _at all_ which model would be "the best" in terms of response quality.

Here's the current state with version numbers as far as I can piece it together (using my best guess at naming of each component of the version identifier. Might be totally wrong tho):

1) prefix (optional): "gpt-", "chatgpt-"

2) family (required): o1, o3, o4, 4o, 3.5, 4, 4.1, 4.5,

3) quality? (optional): "nano", "mini", "pro", "turbo"

4) type (optional): "audio", "search"

5) lifecycle (optional): "preview", "latest"

6) date (optional): 2025-04-14, 2024-05-13, 1106, 0613, 0125, etc (I assume the last ones are a date without a year for 2024?)

7) size (optional): "16k"

Some final combinations of these version number components are as small as 1 ("o3") or as large as 6 ("gpt-4o-mini-search-preview-2024-12-17").

Given this mess, I can't blame people assuming that the "best" model is the one with the "biggest" number, which would rank the model families as: 4.5 (best) > 4.1 > 4 > 4o > o4 > 3.5 > o3 > o1 (worst).

Re: OpenAI dropped the price of o3 by 80%

#382

Post by tethys »

gloosx wrote: post #374
_joel wrote: post #370

Earlier quoted context omitted.

I think modern face verification has moved on, it's been video in all my encounters.

still no real human is involved, as they mention their verification is automated and prohabilistic — which is especially funny to hear in context of verification. Im pretty sure even a kid can go around it, e.g. on the video showing a photo of a person holding his passport which you can find online.

No. You have to turn your head, and stuff. Also, even if this would work, they allow only one verification per person per 90 days.

Re: OpenAI dropped the price of o3 by 80%

#383

Post by rfoo »

behnamoh wrote: post #10

how do we know it's not a quantized version of o3? what's stopping these firms from announcing the full model to perform well on the benchmarks and then gradually quantizing it (first at Q8 so no one notices, then Q6, then Q4, ...). I have a suspicion that's how they were able to get gpt-4-turbo so fast. In practice, I found it inferior to the original GPT-4 but the company probably benchmaxxed the hell out of the tu…

I don't work for OAI so obviously I can't say for them. But we don't do this.

We don't make hobbyist mistakes of randomly YOLO trying various "quantization" methods that only happen after all training and claim it a day, at all. Quantization was done before it went live.

Re: OpenAI dropped the price of o3 by 80%

#384

Post by Squarex »

resource_waste wrote: post #237
lvl155 wrote: post #42

Google has been catching up. Funny how fast this space is evolving. Just a few months ago, it was all about DeepSeek.

Deepseek was exciting because you could download their model. They are seemingly 3rd place and have been since Gemini 2.5.

I would put them on the fourth after Google, OpenAI and Anthropic. Still the best open weight llm.

Re: OpenAI dropped the price of o3 by 80%

#385

Post by sunaookami »

lxgr wrote: post #48

Is there also a corresponding increase in weekly messages for ChatGPT Plus users with o3? In my experience, o4-mini and o4-mini-high are far behind o3 in utility, but since I’m rate-limited for the latter, I end up primarily using the former, which has kind of reinforced the perception that OpenAI’s thinking models are behind the competition altogether.

200 per week now: https://x.com/kevinweil/status/1932565467736027597

Re: OpenAI dropped the price of o3 by 80%

#386

Post by rfoo »

csomar wrote: post #327
tedsanders wrote: post #159

Earlier quoted context omitted.

Nope, not what we’re doing. o3 is still o3 (no nerfing) and o3-pro is new and better than o3. If we were lying about this, it would be really easy to catch us - just run evals. (I work at OpenAI.)

I think the parent-parent poster has explained why we can't trust you (and work on OpenAI doesn't help they way you think it does). I didn't read the ToS, like everyone else, but my guess is that degrading model performance at peak times will be one of the things that can slip through. We are not suggesting you are running a different model but that you are quantizing it so that you can support more people. This can'…

An (arbitrarily) quantized model is a totally different model, compared to the original.

Re: OpenAI dropped the price of o3 by 80%

#388

Post by anticensor »

joshstrange wrote: post #152
Szpadel wrote: post #92

Earlier quoted context omitted.

how do you sample it behind the scenes? usually best of X means you generate X outputs and you choose best result. if you could do this automatically, it would be game changer as you could run top 5 best models in parallel and select best answer every time but it's not practical because you are the bottleneck as you have to read all 5 solutions and compare them

I think the idea is they use another/same model to judge all the results and only return the best one to the user.

I think the idea is they just feed each to the RLHF reward model used to train the model and return the most rewarded answer.

Re: OpenAI dropped the price of o3 by 80%

#389

Post by jstummbillig »

mythz wrote: post #358

I've been turned off with OpenAI and have been actively avoiding using any of their models for a while, luckily this is easy to do given the quality of Sonnet 4 / Gemini Pro 2.5. Although I've always wondered how OpenAI could get away with o3's astronomical pricing, what does o3 do better than any other model to justify their premium cost?

It's just a highly unoptimized space. There is very little market consolidation at this point, everyone is trying things out that lead to wildly different outcomes and processes and costs, even though in the end it's always just a bunch of utf-8 characters. o3 was probably just super expensive to run, and now, apparently, it's not anymore and can beat sonnet/opus 4 on pricing. It's fairly wild.

Re: OpenAI dropped the price of o3 by 80%

#390

Post by cactusplant7374 »

ddtaylor wrote: post #314
34679 wrote: post #226

I'd like to offer a cautionary tale that involves my experience after seeing this post. First, I tried enabling o3 via OpenRouter since I have credits with them already. I was met with the following: "OpenAI requires bringing your own API key to use o3 over the API. Set up here: https://openrouter.ai/settings/integrations " So I decided I would buy some API credits with my OpenAI account. I ponied up $20 and started…

I also am using OpenRouter because OpenAI isn't a great fit for me. I also stopped using OpenAI because they expire your API credits even if you don't use them. Yeah, it's only $10, but I'm not spending another dime with them.

That is so sleezy.
Post reply on HN