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

251–260 of 518 posts

Re: OpenAI dropped the price of o3 by 80%

#251

Post by mliker »

CSMastermind wrote: post #55
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…

This is almost certainly what they're doing and rebranding the original o3 model as "o3-pro"

Where are you getting this information? What basis do you have for making this claim? OpenAI, despite its public drama, is still a massive brand and if this were exposed, would tank the company's reputation. I think making baseless claims like this is dangerous for HN

Re: OpenAI dropped the price of o3 by 80%

#252

Post by linsomniac »

tedsanders wrote: post #208
arresin wrote: post #195

Earlier quoted context omitted.

Not quantized?

Not quantized. Weights are the same. If we did change the model, we'd release it as a new model with a new name in the API (e.g., o3-turbo-2025-06-10). It would be very annoying to API customers if we ever silently changed models, so we never do this [1]. [1] `chatgpt-4o-latest` being an explicit exception

>we'd release it as a new model with a new name

Speaking of a new name. I'll donate the API credits to run a "choose a naming scheme for AI models that isn't confusing AF" for OpenAI.

Re: OpenAI dropped the price of o3 by 80%

#253

Post by BeetleB »

valleyer wrote: post #213
BeetleB wrote: post #160

Why does OpenAI require me to verify my "organization" (which requires my state issued ID) to use o3?

Don't bother anyway. There are lots of cases of people trying and failing to go through the process, and there is no way to try a second time. https://community.openai.com/t/session-expired-verify-organi... https://community.openai.com/t/callback-from-persona-id-chec... https://community.openai.com/t/verification-issue-on-second-... https://community.openai.com/t/verification-not-working-and-... https://community.ope…

Yikes! Indeed, I won't bother.

Re: OpenAI dropped the price of o3 by 80%

#254

Post by xmprt »

amanda99 wrote: post #38
xmprt wrote: post #29

Earlier quoted context omitted.

How are LLMs cached? Every prompt would be different so it's not clear how that would work. Unless you're talking about caching the model weights...

You would use a KV cache to cache a significant chunk of the inference work.

Using KV in the caching context is a bit confusing because it usually means key-value in the storage sense of the word (like Redis), but for LLMs, it means the key and value tensors. So IIUC, the cache will store the results of the K and V matrix multiplications for a given prompt and the only computation that needs to be done is the Q and attention calculations.

Re: OpenAI dropped the price of o3 by 80%

#255

Post by az226 »

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…

Even classic GPT-4 from March 2023 was quantized to 4.5 bits.

Re: OpenAI dropped the price of o3 by 80%

#256

Post by az226 »

dist-epoch wrote: post #149
hyperknot wrote: post #128

Earlier quoted context omitted.

I got 700+ tokens/sec on o3 after the announcement, I suspect it's very much a quantized version. https://x.com/hyperknot/status/1932476190608036243

Or maybe they just brought online much faster much cheaper hardware.

Or they are using a speedy add-on decoder.

Re: OpenAI dropped the price of o3 by 80%

#257

Post by lossolo »

simonw wrote: post #158
bitpush wrote: post #62

Earlier quoted context omitted.

The problem is your costs also scale with revenue. Ideally you want to have control costs as you scale (the first you build is expensive, but as you make more your costs come down). For OpenAI, the more people use the product, the same you spend on compute unless they can supplement it with another ways of generating revenue. I dont unfortunately think OpenAI will be able to hit sustained profitability (see Netflix f…

"... as you make more your costs come down" I'd say dropping the price of o3 by 80% due to "engineers optimizing inferencing" is a strong sign that they're doing exactly that.

> "engineers optimizing inferencing"

They finally implemented DeepSeek open source methods for fast inference?

Re: OpenAI dropped the price of o3 by 80%

#258

Post by stevage »

34679 wrote: post #242
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…

Follow-up. I pasted the above into ChatGPT because I genuinely wanted to know how to get my 20 bucks back. That request is in and they say I'll get a response in 2-3 days. I then asked it "What are the potential risks of agreeing to their terms?" Here's the response: Agreeing to Persona’s terms, especially for biometric identity verification, involves both privacy and long-term data security risks. Here’s a clear bre…

Huh, I never considered using ChatGPT to read terms of serivce for me. Nice idea.

Re: OpenAI dropped the price of o3 by 80%

#259

Post by beering »

mliker wrote: post #251
CSMastermind wrote: post #55

Earlier quoted context omitted.

This is almost certainly what they're doing and rebranding the original o3 model as "o3-pro"

Where are you getting this information? What basis do you have for making this claim? OpenAI, despite its public drama, is still a massive brand and if this were exposed, would tank the company's reputation. I think making baseless claims like this is dangerous for HN

I think Gell-Mann amnesia happens here too, where you can see how wrong HN comments are on a topic you know deeply, but then forget about that when reading the comments on another topic.

Re: OpenAI dropped the price of o3 by 80%

#260

Post by beering »

lispisok wrote: post #18
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 swear every time a new model is released it's great at first but then performance gets worse over time. I figured they were fine-tuning it to get rid of bad output which also nerfed the really good output. Now I'm wondering if they were quantizing it.

It’s easy to measure the models getting worse, so you should be suspicious that nobody who claims this has scientific evidence to back it up.
Post reply on HN