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

Advancing the price-performance frontier with GPT‑5.6

openai.com

Post reply on HN
Print view View original

211–220 of 424 posts

Re: Advancing the price-performance frontier with GPT‑5.6

#211

Post by archon1410 »

If Luna is so good and cheap, I'm wondering why ChatGPT Free users still only have access to GPT 5.5 Instant. The API pricing for chat-latest is the same as Sol(!),[1] while Luna is 25× cheaper [2]. Even assuming highly inflated API pricing for chat-latest, Luna not being the daily driver on Free and Go plans feels off.

Perhaps users prefer instant responses over thinking models so much so that using a more expensive and less performant non-thinking model is worthwhile.

[1] https://developers.openai.com/api/docs/models/chat-latest [2] https://developers.openai.com/api/docs/models/gpt-5.6-luna

Re: Advancing the price-performance frontier with GPT‑5.6

#212

Post by kridsdale1 »

re-thc wrote: post #116
jpadkins wrote: post #86

Earlier quoted context omitted.

When model intelligence reliably hits 90%-95% of current day knowledge worker tasks, they are going to burn those weight to silicon and we will see another 10X improvement in price/performance frontier. The dynamic GPU clusters will be used for the 5% of tasks, and pushing out the frontier. Also there will be a set of knowledge tasks that are not done today (because they are too difficult for most knowledge workers),…

> When model intelligence reliably hits 90%-95% of current day knowledge worker tasks, they are going to burn those weight to silicon Google is already working on a similar idea but more "flexible".

Explain.

Re: Advancing the price-performance frontier with GPT‑5.6

#214

Post by FuriouslyAdrift »

jrflo wrote: post #115
jpadkins wrote: post #86

Earlier quoted context omitted.

When model intelligence reliably hits 90%-95% of current day knowledge worker tasks, they are going to burn those weight to silicon and we will see another 10X improvement in price/performance frontier. The dynamic GPU clusters will be used for the 5% of tasks, and pushing out the frontier. Also there will be a set of knowledge tasks that are not done today (because they are too difficult for most knowledge workers),…

Burning the weights into silicon would be many orders of magnitude increase, not just 10x. It's kind of crazy that this hockey stick the AI hype bros talk about seems more and more every day like it might be real

Yep it will be ASICs and DSPs all over again. Orders of magnitude changes.

Re: Advancing the price-performance frontier with GPT‑5.6

#215

Post by kridsdale1 »

dannyw wrote: post #120
Imanari wrote: post #110

Earlier quoted context omitted.

How do you run 'deep research'?

Deep research is basically a LLM with web search, and a "work really hard" goal-orientated prompt, and some output formatting suggestions.

And self forking fan out.

Re: Advancing the price-performance frontier with GPT‑5.6

#216

Post by njcornell »

in_a_society wrote: post #142
GodelNumbering wrote: post #44

"Half the money I spend on advertising is wasted; the trouble is I don't know which half." -John Wanamaker This applies even more strongly to model choosing. I know for a fact that majority of my work doesn't require a very strong model, but separating the trivial and non-trivial tasks is a famously hard problem (if at all decidable).

I don't see why it should be all that difficult. All you have to do is first find a library that implements a decent solution to the halting problem and you're off to the races.

You can use my script p-noteq-np.sh too if that helps.

Re: Advancing the price-performance frontier with GPT‑5.6

#217

Post by simianwords »

Decabytes wrote: post #188

Has anyone ever done a comparison between the smaller models like Luna, against the previous GPT 5 frontier models? Have we gotten to the point where the small models are as good as the frontier models of the past, or is there still a way to go?

I ran a prompt with ChatGPT since I'm also curious. The price reduction is crazy.

- GPT-5 high: score 35, approximately $0.37/task

- Luna medium: score 38, approximately $0.01/task

- Luna max: score 51, approximately $0.042/task

So Luna medium is:

- slightly more capable than GPT-5 high;

- approximately 35–40× cheaper per benchmark task.

And Luna max is:

- 16 Intelligence Index points better;

- still roughly 9× cheaper per task.

This reduction was possible within 1 year.

Re: Advancing the price-performance frontier with GPT‑5.6

#218

Post by kridsdale1 »

handfuloflight wrote: post #89
bob1029 wrote: post #73

Earlier quoted context omitted.

Anything related to reading and interpreting the environment seems to always benefit from the addition of more agents to the search party, assuming you have some rational way to synthesize their results. Taking actions that mutate the environment is a different story. I think this is where you run into diminishing returns very quickly. You generally want one strong agent to act given the results of all the searching…

I definitely think you want the genius model to synthesize everything that rolls up to them.

This is why in your brain you have trillion threads processing and summarizing sensor data (immutable functions), but a SINGLE thread of “execution” which we call the conscious soul.

Re: Advancing the price-performance frontier with GPT‑5.6

#219

Post by dgellow »

paxys wrote: post #206
dgellow wrote: post #198

Earlier quoted context omitted.

But they still have to cover compute cost, and they already committed to more than $650B in infra expenses for 2035

Why assume they are not already making up the compute costs for smaller models?

Enough for such a massive reduction? If yes that’s really impressive

Re: Advancing the price-performance frontier with GPT‑5.6

#220

Post by andai »

handfuloflight wrote: post #89
bob1029 wrote: post #73

Earlier quoted context omitted.

Anything related to reading and interpreting the environment seems to always benefit from the addition of more agents to the search party, assuming you have some rational way to synthesize their results. Taking actions that mutate the environment is a different story. I think this is where you run into diminishing returns very quickly. You generally want one strong agent to act given the results of all the searching…

I definitely think you want the genius model to synthesize everything that rolls up to them.

I think this is an unsolved problem. The most interesting thing I saw here is the Recursive Language Models paper.

https://arxiv.org/abs/2512.24601

There's also a great write up here by the author:

https://alexzhang13.github.io/blog/2025/rlm/

Post reply on HN