ZH version is available. Content is displayed in original English for accuracy.
Advertisement
Advertisement
⚡ Community Insights
Discussion Sentiment
79% Positive
Analyzed from 12362 words in the discussion.
Trending Topics
#sol#luna#gpt#astra#codex#usage#more#model#models#claude
Discussion (447 Comments)Read Original on HackerNews
Here's GPT-6 Luna pelicans: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...
And GPT-6 Sol: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...
Scroll to the bottom for the GPT-6 Sol max one: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...
For comparison, here are the pelicans I got for GPT-6 Astra: https://tools.simonwillison.net/markdown-svg-renderer?url=ht... - I still like the Astra Max one best.
Here's a comparison grid showing all of the GPT-6 and GPT-5.6 pelicans at all effort levels: https://static.simonwillison.net/static/2026/gpt-6-and-5.6.h...
The grid is actually really interesting, because it shows that the 5.6 family default to brighter colors than the 6 family.
Snark aside, Luna 5.6 was (is) an incredible game-changer.
perhaps it then does mean - squeeze as much as you can get off this actual free usage.
By raising it from investors.
https://en.wikipedia.org/wiki/Pareto_front
https://en.wikipedia.org/wiki/Pareto_principle
I assume it's a subsidy to get more training data.
EDIT: Okay downvoters, what's your take on why they're giving away Luna for so cheap?
I had all the tabs open individually and harder to scan which model is which... otherwise keep up the great work! I like the grid view a lot. (Also the pages have no OG images set, which impacts what the link looks like shared)...
OG images will require me to move away from publishing in a Gist and linking to from a JavaScript page that loads the Gist. Probably worthwhile though.
1 - https://bench.killswitch-lang.org
Is it? It was already too cheap to meter for me. Luna 6 is actually worse on some benchmarks than 5.6. I’d have loved improved performance for 2x the price than ~equal performance for 0.5x the price.
I expected a Fable 5 -> Opus 5 situation, where GPT 6 Sol would perform on par with GPT 6 Astra.
Instead it's more like a price cut on GPT 5.6 Sol, and I'll have to stick with Astra for my work.
The only thing I can hope for is that more users switching to the GPT 6 Sol model frees capacity, allowing OpenAI to hand out some usage resets.
They put both legs on the same side of the bike.
Even Astra max which actually put one leg on each side of the bike still somehow messed it up because when it added the bike chain, it put the left leg between the bike chain and the frame.
Half the price when it launched, or after the price dropped by 75%?
> GPT‑6 Luna vs. GPT‑5.6 Luna | $0.20 → $0.10 | $1.20 → $0.50 | 50% cheaper
I can read it as follows (below), meaning that GPT-5.6 is 50% cheaper.
- GPT-6 = $0.20
- GPT-5.6 = $0.10
good god
I'm so tired of looking at benchmarks. I always look fwd to the pelicans.
And Astra medium seems to yield similar or better quality for the same price as Sol 6 xhigh.
1/ Usage limits: downstream of input/output cost, but resets and obscure windows and odd 20x plan / 5x plan != 4x usage math throw a wrench into it. Winner right now is Codex by a mile, especially when you factor in the fact that ChatGPT usage (even 6 Astra Pro) is essentially unmetered on the 20x plan. Always a bummer when asking if I should see a doctor about a rash means I can't code as much. It's also is a godsend if you use an MCP like oracle to automate the process of calling the Pro model on particularly tough problems, giving better planning results or deeper code analysis without burning usage.
2/ Context window in the harness. Claude Code wins on this. There used to be a toml file workaround for Codex to extend the GPT context window to 1m, but this stopped working on the plans and only on per-token billing (ETA: noname120 pointed out this is no longer the case and it can be enabled again [1]). 252k is just not enough. Codex's compaction is very good, fwiw, but it happens so frequently that even a model as powerful as Astra sometimes loses the plot on long-running tasks.
3/ Ability to use the plan outside of the official harness. Codex wins. Anthropic does shit like bills requests as extra usage if it sees a hermes.md in a commit.
I've subscription hopped a bunch, and at times I've had both, but I keep coming back to Codex because it wins on 2/3.
ETA: apparently I haven't been Keeping Up With the Altmans and new 20x signups have been disabled for a few weeks. I am grandfathered in, which makes the comparison above pretty much moot.
[1]https://news.ycombinator.com/item?id=49806060
This hasn't been the case since around July. If you measure usage in raw api costs, Anthropic is actually giving more on $200 than OpenAI now. This includes resets. Usage allocation difference would be humiliating for codex subs were it not for resets. But fixing usage limits with resets is ugly, and they're not good for your mental well-being.
> Context window in the harness
Codex now allows 1M for subs with config params. But generally speaking, you shouldn't really be using 1M context. If you accidentally send a request with say, ~700k context already accumulated in a session which is outside cache TTL, you're paying full cost of these 700k tokens.
> I've subscription hopped a bunch
OpenAI actually has a new strategy to prevent subscription hopping after their 2-3 month-long marketing push to get claude-folks to switch over:
you can't buy a $200 sub anymore. So if you cancel, you won't be able to get back in. Hostage situation, essentially.
EDIT: re: usage limits, oh-my-pi maintainer has been tracking this - https://nitter.xitter.cc/_can1357/status/2090075496948060372
I think I’m gonna move back to a Claude plan. I could barely hit the $200 limit if I went non-stop on programming tasks.
When they started the aggressive campaign, entire X (including myself, sadly) was full of posts about how "unlimited" codex usage is even on a $20 plan. Sam Altman was posting something in line of "we love our users, unlike Anthropic". Got my network to get codex subs because of the value compared to claude.
Then they gradually reduced the limits to the point where even $200 plan only lasts you just 1-2 days and $20 is basically unusable, then the hostage thing.
Do you have /fast enabled by any chance?
Opus is at least actually usable even on the small plan. The main downside is its insane writing style, but 5.5 seems to address that somewhat. Otherwise, you can just use your $20 OpenAI plan to have Luna de-slop Opus' prose, which seems to work fine.
I haven't been tracking, but this roughly matches my experience with codex 20x and claude 20x subs. Claude subscription now lasts me 3-3.5 days on average. Codex is 2-2.5 days. This is work on same projects, with similarly sized tasks.
To make matters worse, I've merged a lot more code produced by fable than sol/astra.
But even if we leave that aside, OpenAI models are also much more eager than Anthropic, which are on the lazier side. Left unsupervised, Sol/Astra will attempt to build a sha256 verified rocket ship if you ask them to fix a race condition in your to-do list app. Anthropic models will do what you asked for, maybe even forget to implement parts of that ask, but they won't generally throw a slop granade at you.
I can leave Fable orchestrator unsupervised for ~2h. Leaving Sol/Astra unsupervised for ~2h means the next user turn will contain a message: "what are you doing and why?".
Are you sure?
https://x.com/thsottiaux/status/2098113585683808624
While you're understandably not including the values of the $20 standard plans on both, I find the generosity of then token limits on ChatGPT plus vs Claude Pro (it's a huge difference) to be good representation of their respective attitudes towards the average user. You literally cannot use Claude pro to build real software, unless you're extremely frugal with your prompts and don't try anything even a little ambitious.
Also, Anthropic has zero models comparable to Luna.
They're both pretty horrible, but I find it difficult to find arguments for why Anthropic is worse than OpenAI, other than their doomtrolling. Which, in the grand scheme of things, doesn't even register.
Edit: forgot about the SpaceX thing.
Anthropic is trying to kill open models way harder
That alone is reason enough. Also, I don't think either of them are horrible. That's honestly a ridiculous take considering how much people in here love their models, and how much they've advanced the industry forward.
Anthropic does all that but they're also populated by many people who believe they are building God and that they must build their god first in their own image so that it can take control of humanity and protect us from any competing god which is not built in their image. Their position is inherently paternalistic and authoritarian, and they consider suppression of competition not just important to the bottom line but to life in the universe. Under the doomer ethos there is no evil too great to rationalize.
There are plenty of wrongs done in the name of profit, but capitalists have nothing on zealots in terms of causing serious harm. Profit motives can be directed by influencing incentives, but zealotry is frequently terminal.
That isn't to say that there isn't some overlap-- the cultists have infected both organizations. But OpenAI has pretty consistently only given lip service to AI doom to the extent that it improves the bottom line, while (mis)Anthropic was founded specifically because OpenAI wasn't mentally ill enough.
> Also, OpenAI is just a company I'd rather support than Anthropic.
Sure, he's free to say whatever especially considering the amount of revenue he's creating, but it's just an altitude that I prefer not to see.
As far as I know Codex (at least the GUI) can automatically call the ChatGPT Chat models (including Astra 6 Pro), you just need to @ a ChatGPT Chat conversation from within Codex and tell it when to use it.
> There used to be a toml file workaround for Codex to extend the GPT context window to 1m, but this stopped working on the plans and only on per-token billing
Not true, it works again[1]. I confirm that it works both on 5.6 Sol and Astra 6, possibly other models too.
[1] https://x.com/thsottiaux/status/2089082893804896524
And re: the toml workaround, AWESOME! I appreciate you pointing these two things out, this is my highest-ROI HN comment thus far.
I'm quite puzzled about why Anthropic is so hellbent on blocking other coding agents. It's not like Claude Code has any secret sauce, right? And doesn't Anthropic make monkey off API usage, and their magic is on the model side anyway?
But to be fair, they don't really enforce the harness rule that much anymore. I guess if your harness doesn't do a lot of weird things like a lot of cache misses, or triggers some distillation attacks, or some broader Chinese fingerprints, they're tongue-in-cheek okay with you using a third party harness.
Having limitless webUI ChatGPT usage is much better user experience, though. I'll give them that.
(edit: Sol-6 is half the price, so maybe the usage limits are going to be way better.)
Meanwhile I just burned ~20% of my weekly quota with Astra making one config file for a service.
I appreciate and follow Matt Pocock's advice: avoid autocompaction. Compaction is lossy, which is ok when you're managing it at phase boundaries, but autocompact is lossy at the most inopportune times, firing mid-task and leading to agents going off the rails.
My conversations compact hundreds of times. By the time it has done a dozen or so compactions, it fully understands the work I want it to do (and how). It's almost like having a fine-tuned Astra model.
10/10, would recommend.
The GPT was about 1B on two projects on 300$ worth of plans all on Astra and I capped out on usage.
Anthropic caching must be better because the cache rates are better on Claude models.
Compare that to Claude and I can run multiple agents on Opus almost indefinitely. YMMV of course but I was shocked at how quickly I burned through Codex usage.
On the context window, I feel so cramped on Codex, compacting happening every time I turn around is annoying. I didn't realize how much I enjoyed the Claude context window size.
Makes me think they picked Codex, stopped trying Claude, and just hang on to outdated beliefs about the value they're receiving.
Opposite in my experience. I need to limit codex to 500k on medium/low, still run out in 2-3 days with 1 CLI window. CC gives me 4-5 medium/high days with 2-3 CLI windows, and Opus is still great for other regular dumb engineering/refactoring.
On the other hand my head starts to hurt if I read Opus for too long, hopefully they fixed it with 5.5.
Using Agentsview (which might have it's own issues) I was getting ~$200 of API usage in my 1 week Codex window (paid $100) vs ~$5,000 of API usage in 1 week for Claude (paid $200).
And ya i can go over that 240k limit, I still very seldom do, and try to treat it as the actual limit. I'm surprised to see so many people still talking about compaction to complete long running tasks, i think the bulk of the work should be somewhat frontloaded into a plan that is split off into subplans, then you can kinda open up a few options, one session with subagents for the subplans of the main plan, or just handoff prompts about progress against the main plan/relevant subplan. I just never trust the blackbox that is compaction, I feel its a recipe for disaster/context poison.
I’m currently on the 5x plan and burned through 5% today on a difficult task in 15 minutes so I doubt that. If you got the wrong kind of tasks that you work on, it can go fast.
LMAO, I wish this were true, I hit limits (and the "we are disabling access to protect your data" warnings) all the time, or have chats just...fuck off and get into weird/invalid states (interrupted chats, chats that are spinning and stuck, returning "/mnt/" paths instead of images/md files, file links being returned with no file backing them, image classifier firing...and then returning the image anyway (though now I know that GPT-Image-X really really wants to generate NSFW even when that isn't the request)).
Though I am probably an outlier, I have both 20x Claude/ChatGPT plans and max both out every week, so... (in my defense I am a hobbyist and this is out-of-pocket)
Is this not the default anymore? I am on the (now closed) 20x plan.
That isn't a valid comparison, since Codex 20x is closed. So we should be comparing Claud 20x to Codex 5x + credits.
Also in Codex, even though you can increase the context window to 1m so its on par with Claude, exceeding the default is billed at 2x.
Like you said, it felt very natural to work with. Opus 5 is way too slow and verbose for me, I find I get distracted and annoyed with it.
Opus 5.5 seems a LOT closer so far to what I liked about 5.6 Sol but we'll see
It wants things beyond what the mortals (us) know to reach for. It's not good at explaining itself, it doesn't show it's thinking. It's often not wrong. But the no compromises attitude can be unbearable to deal with. Especially given how little it cares about telling us.
Otherwise I'm using 5.6 Sol for actual plan execution and review..
Astra was/is superior for planning type tasks. It was capable of doing seemingly magic things with rather vague/lazy instructions ("I need to be able to test this on Windows, maybe a qemu VM or something? Shrug." ... 1 hour later "yeah i built you a whole qemu + eval windows image + harness of powershell scripts + shell scripts to retrieve & verify harness.").
And for UI work -- which is not something I do a lot of but do here and there -- it was clearly superior to 5.6 Sol.
But it also feels sloppier? Somehow. And too expensive to use.
We'll see how Sol 6 is.
Terra had the "workhorse" quality where it could do these changes in bulk and follow directions without being too 'smart' (but sloppy) as you described. Luna was a bit too dumb and would make sloppy mistakes; I see that more as a "run these tests and format the results" sort of model. Maybe 6 Luna will be better.
I also just reread your comment and realized the naming convention is still extremely confusing with respect to ordering of [Family]x[Model]x[Number].
If I leave Astra overnight, I'll wake up with three new different projects, each of them 20% done and having nothing to do with my original goal.
Highlight and lowlight of my week was successfully convincing the OpenAI support chat robot to give me a refund for the month for my issues with 6 chewing threw my usage with no output.
Input
Output
Price reduction
GPT‑6 Sol vs. GPT‑5.6 Sol
$4 → $2
$20 → $10
50% cheaper
GPT‑6 Luna vs. GPT‑5.6 Luna
$0.20 → $0.10
$1.20 → $0.50
50% cheaper
The parent + subagent workflow has become critical for keeping the reasoning agent (parent) context-lean while also letting me chat to the main agent while work is getting done.
My main process is to use Fable to reason and then spawn Opus subagents, and I get amazing results, and I'm always looking into what the subagents are doing.
- Unbearably slow
- A token eating machine like no other
- Constantly compacting
- A model (like other GPT ones) that hides thinking traces and thinking summaries, which infuriates me
I've been in the Claude camp for a while, but the way it writes has left me with a a brick for a brain and wanted to see if Astra was as good as they say. Well, I can't know, because in the time it takes for it to actually build anything useful, I've moved to other ideas.
Unbearably, annoyingly slow. I keep thinking I must be doing something wrong.
However, it is not a 'token eating machine'. In fact it uses a third of the output tokens of Opus 5.5, Fable 5.1, or Opus 5.
17k for Astra xhigh vs 61-66k.
https://developers.openai.com/api/docs/pricing?latest-pricin...
>> $4 → $2
>> $20 → $10
Do you mean 100% more expensive? GPT 6 is 100% more expensive than 5.6 per your post.
Its a great release, I will use both heavily.
This is great, but practically, I'm not going to start working on more side projects.
Perhaps in another 6-12 months I'll be fine to drop down to $20/m instead of $200.
A lot of what I'm doing has pretty expensive build/testing processes between iterations - even on a 40 core machine - so I'm not burning tokens 24/7 like some people may.
I'd guess I'm probably spending >50% of the time running tests & build processes & tooling and the remainder is purely burning tokens.
I also have some internal tooling (that I will hopefully open source soon) that makes LLMs substantially more correct (thus more efficient) - so there's that, too.
There are a ton of use cases that open up with cheaper models.
E.g. extensive security scanning on every PR, quality scans, adversarial reviews etc
If these price changes mean that coding plans have effectively more usage then that's great, but Codex is surviving on resets from my own experience using it. I was glad to go back to Claude.
Should be B vs A correct?
Else it's confusing
We don't know how much they are bleeding financially, it might just be a front
If they’re subsidizing my usage, that’s great.
When they cut prices on luna the first time around they took (literally) millions of users from anthropic.
The "paradox" is when an increase in efficiency which would decrease the use of a resource all else equal, instead indirectly causes more use.
Highly subjective take
What kind of work do you do, out of curiosity
Cache read/write decrease by 50% or similar? That's where most (95%+) of the cost is for agentic coding workloads.
https://developers.openai.com/api/docs/pricing
I actually preferred 5.6-Terra not because it is technically superior (it isn't) but because it had better instincts to NOT do this stuff.
PS - Speaking of better instincts, have they closed the UI-design gap at all? I keep a Claude subscription just because /design produces significantly higher quality UI design/UI feedback/UI refinement than anything I've seen from OpenAI.
IMO 5.6 Sol had this weird dead zone between medium and high where medium under engineered and took short cuts and high over engineered and ignored instructions it didn't agree under the guise of trying being helpful. The whole 5.6 line was the first release from OpenAI where it felt like reasoning level really mattered and was incredibly finicky.
I haven't felt similar issues with GPT 6 though and am very happy with Astra low/med/high as my default choices depending on the task.
In general, I felt like with 5.6 the effort level did less than previous to make the models smarter and more just increased the complexity of the response. I have a half joke theory based only on vibes that OpenAI splitting 5.6 into Sol/Terra/Luna is where the intelligence split happened and so the effort levels were just like "think harder about the decision you already made". So like if the model decided the earth was flat on low effort it'd just say something like "the earth is flat because the horizon is flat". If it was on xhigh reasoning it'd give you a massively complex answer about how the sun reflects light because of the ozone layer and why people flying in planes can see a curve. In both cases though, adding more effort wouldn't get it to realize the earth was round. It just made the answer about it being flat more complex.
To be clear, that theory is not meant to be taken too seriously. It's not based on anything other than vibes. It's just my way of explaining to myself something I'm frustrated about to myself.
The GPT / Codex models have always been "overengineer" personalities. I prefer that to "I left a pile of race conditions lying around and big gaps in testing" though, which is what I was getting from Opus at times.
But yes both Astra and Sol veer on the side of paranoid. And honestly that's better for team work. For solo work where you just want to yeet something, it can be tiring.
You learn to tame the GPT "personality" on this front by combing over once a week and asking it to find and exterminate pointless tests, clean abstractions etc.
All I want to know is how old is the model and how much does it cost. I can figure out which one I want to use based on that, assuming that newer models are always better.
Trying to convince us there is a difference between GPT-6-Sol and GPT-5.6-Terra or whatnot is ludicrous to the point of being insulting, especially when new models come out every week.
If you have something that needs to be done right, might be a bit complicated, up the model size.
You can see this in the pelicans. Big model pelicans are pretty accurate by default. Up the reasoning and only more so, but with more detail. For Astra, it is 105 lines for low, 250 lines for max reasoning.
Small model pelicans will lack the fidelity of a large model. Bits will be out of place etc. For luna, it's 90 lines for low, 150 lines for xhigh.
Additionally the amount of time taken is increased for the larger models. Luna takes 11 seconds on low, and 1:33 for xhigh. Astra is 33 seconds on low, 4 minutes on max.
And naturally, there is the cost. There's some overlap in functionality between luna xhigh and Astra low in the sense that luna really can do quite a suitable job for some tasks. But there are just some tasks that just don't make sense for Luna, even at high reasoning.
The other thing to remember is that sometimes high fidelity isn't ideal. It can lead to overdesigning. My recommendation is to commit early, commit often, and review everything you do, which we've all been doing since before LLMs right?
https://developers.openai.com/api/docs/guides/reasoning?api-...
But effort is basically how much extra internal scratchpad to use and how much extra questions to ask and answer before producing a result, Exploring more hypotheses, validating consistencies, calling more tools.
If you’re happy with your token spend on Astra then keep doing what you’re doing. but if you feel the need to conserve tokens, then you can do that by switching to smaller models like Luna when the task is straight forward.
EDIT: this doesn't say anything about availability on either Azure or AWS. I'm assuming it will show up later, but it would be interesting if it didn't.
Edit: Yes, it applies also to subscriptions, source https://x.com/thsottiaux/status/2102463847714247142
Design: https://image.non.io/78795662-8bfc-4e14-8d72-3738392aa6b3.we...
All 3 were given the same prompt to dynamically light these and to create the designs as a SPA with page transitions.
Astra: https://html.non.io/annui-astra
Sol: https://html.non.io/annui-sol
Luna: https://html.non.io/annui-luna
Luna gets the button wrong, and in the same way Grok/MiMo did. Looking into it more, it's because Luna actually searched my computer for similar builds, found the ones that I did for grok/mimo, and referenced their files. Astra is still the best by a significant margin in my eyes. Far more polish, better page transitions, effects that aren't overcooked and take into account the page. Better contrast.
Now GPT 6 Luna is even cheaper, and more intelligent, there is no going back... to SOL 5.6 for intelligent layer.
I was hoping for a serious Luna upgrade. It was already cheap enough. This feels more like a price reduction than an upgrade.
That said, if the new Luna is able to handle ultra mode and subagents v2 in codex cli, then at least that’s a win.
I forgot which model degraded in quality as time went by, but let's try out Luna 6 for a few more days to confirm for upgradability.
Does anyone know how exactly these price differences for example between sol6 and sol5.6 translate to codex percentages? In theory it seems like for "high" on both it should result in ~3x more usage. If that is actually the case it would be huge! But all we see is % left and % changes while using and we really have no idea when or how those numbers are being calculated or when they change. So there is a 50% price reduction on API but who knows how the hell that translates to whatever price calculation is used on codex.
> In the Coding Agent Index, Sol improves but Luna regresses: In OpenAI's Codex harness, GPT-6 Sol (max) scores 57 in the Artificial Analysis Coding Agent Index, up 2 points from GPT-5.6 Sol (max), with gains in Terminal-Bench 4.0 (43% vs 37%) and SWE-Atlas-QnA (58% vs 54%). At $2.99 per task it costs ~50% less than GPT-5.6 Sol (max) and sits on the Pareto frontier of Coding Agent Index vs Cost per Task. GPT-6 Luna (max) scores 41, down 2 points from GPT-5.6 Luna (max), with lower scores in SWE-Atlas-QnA (44% vs 49%) and DeepSWE v1.1 (64% vs 66%), at ~60% lower cost per task.
https://x.com/ArtificialAnlys/status/2102462962758033624
Given that they had to discontinue sales of the 20x Pro plan after the Astra release due to compute constraints, I wonder if 6 Sol & Luna are smaller vs their 5.6 counterparts?
At least it sounds good on paper, the the graphed results do give me pause as it seems the lower cost might come from a slightly nerfed base model combined with more thinking, going by the more erratic scoring curves and the lower no thinking baseline score. I’ll have to try it out but I really hope they haven’t nerfed Luna/Sol to make this price point possible!
Then they do a new model launch, issue quota resets all around, and it's a party for 2-3 weeks before things return to normal.
I guess the chinese competition spooked them.
Has anyone else noticed this?
Overall, I expect for most people think the winner of today was Anthropic. I personally am preferring Opus 5.5 at medium over GPT-6 Sol Max, in very very early tests. Similar price range, more capability.
But competiton is great, these are solid releases by OpenAI today.
For example, I had Fable review Astra’s output yesterday, and it found some issues and fixed them. Passing the fixes back, Astra then uncovered additional issues with Fable’s fixes (and yes, this will go on ad infinitum if you let it, but these were “real” issues).
It seems the big story here is the reduced Luna pricing. It’s a fantastic model that can handle most automation needs (though I still use the big models for day-to-day development).
Astra is the best. Luna is cheapest then it seems like Sol is the middle child like Terra.
Wierd!!
* Prompt caching dashboard: https://platform.openai.com/usage?usage_section=prompt-cachi...
* Adjust reasoning effort and tool availability without breaking cache
I kind of hated Astra for it's poor instruction following and stopping all the time plus bad code quality. It somehow feels a bit like some of the popular open models but with a lot more knowledge or peek capability. But it doesn't reach peek that often
How do you decide what to pick? I mean, I do Platform work on a large monorepo with many different interconnected services, and so I always want the implementation to be "correct".
> GPT-5.6-Sol is retiring. This conversation will automatically switch to GPT-6-Sol
I don't recall OAI retiring a model so early lol. Similar arch?
OpenAI seems really competitive in most areas, and extremely competitive on cost, but still behind on coding.
Have the frontier labs stopped trying to increase context window size?
Funny thing is they very recently also set a real limit per-user/month, so why even limit the models because theyre "too expensive".
Hardly enough time had passed to develop the data to come to a conclusion. Users can take time to build interest.
Terra ended up just being an awkward middle ground that was not particularly suited for any workload.
Maybe for one-shotting large things Sol is better, but for prod code where I decompose into smaller tasks and read all the code I favored Terra.
Sol 5.6 was still king for architecture/research in my workflow, though.
Hardly enough time had passed to develop the data to come to a conclusion. Users can take time to develop an interest.
Incredible.
Does anyone care about code quality anymore?
This will make me more valuable in the future when everyone has lost the ability to do anything on their own.
Not exactly a low-friction user experience!
Like are you supposed to just somehow intuit, “ah yeah, this task is definitely a GPT-6 Sol Medium task,” or something?
Is this just second nature for OpenAI employees? How are end users supposed to know how to optimally choose a model for a given task? Am I missing something completely here?
Maybe OpenAI can offer an "auto" mode for Codex on the subscriptions, while leaving the possibility of users manually overriding whatever model the router chooses. To me that would be the best of both worlds. The problem is building a competent model router.
Which... fine, I'll take that.
From around GPT 4 results got "Good enough"...I generally try to explain what problem I'm trying to solve, set limitations and boundaries, tell it to ask me questions, have it write up a plan with steps then we take one step at a time.
These new models are starting to feel like iPhone releases where the improvements / feature set feels incremental.
Same on the Claude side which I use for work
I wouldn't be curious to sign up to codex whatsoever these days
These token reset shenanigans are insane
OpenAI is promising "the Sun, the Moon, and the Stars". The spirit of P.T. Barnum is doubtless looking on with jaw dropped at what is beyond doubt one of the greatest demonstrations of chutzpah, by some of the greatest hucksters, in the history of the human race.
Asking cuz I don't think I'm a bot [pats self], I legitimately prefer the GPT models to Anthropic's, don't like Anthropic's customer service/reliability story at all, and I welcome a massive price reduction. Seems like something I should be happy to get.
If you'd told me I'd be typing this a year ago I'd be skeptical though.
But the reason people say "Claude can't compete" is because Claude Opus has been going downhill since 4.7, and many have found Opus 5 intolerable. Fable is much better, but also much more expensive than OpenAI's offerings.
As for any cost based argument, it is immediately invalid because the cost is something that OpenAI fully controls and manipulates.
My previous comment--which suggested astroturfing--was the highest upvoted comment here until it got flagged. Which implies to me that atleast the other remaining humans on this forum see it as well.
26 minutes, 89 comments, upvoted instantly to the top, posted within one hour of the Opus 5.5 announcement. You tell me.
On the other hand, you have previously written: I'll gladly admit I think what these companies are doing is unethical, and I'm sure that biases my thinking toward skepticism. [1]
You have now posted accusations/assumptions of astroturfing and manipulation at least 15 times, without ever providing any evidence. This is in breach of the guidelines, because comments like this poison discussions far more than the comments they're complaining about.
We – of course – want all comments and posts on HN to be authentic. HN is only a place where anyone wants to participate because since the beginning, we've had software mechanisms and moderation practices that detect and weed out inauthentic commenting and voting. We're identifying and dealing with it every day, continually improving the software to detect and remove it. Most of that happens quietly and efficiently in the background without anyone having to see it. When users see evidence of manipulation and report it to us via email, we happily and thoroughly investigate it.
Most of the time, what we find is simply that people are authentically excited and passionate about the topic, which is what is happening here. I understand it can be hard to accept that if you're skeptical about the topic.
It's fine to be skeptical about the topic and you're welcome to express your skeptical views on the topic. People do that every day on HN, about AI-related topics and countless others. Healthy debate is what we're here for.
But you can't keep poisoning HN, by (1) continually posting these unfounded claims, then (2) when users and moderators simply uphold the guidelines, staging a protest by demanding your account be deleted. This is not what people do when they care about a forum's health.
[1] https://news.ycombinator.com/item?id=48220908
> You have now posted accusations/assumptions of astroturfing and manipulation at least 15 times, without ever providing any evidence.
I tried to point out the upvote speed and age-to-comment ratio for this thread look anomalous to me, and that it being posted within the Opus 5.5 release hour was further reason for skepticism. Circumstantial, sure, but I see very little ways to gather hard evidence of astroturfing without being a mod.
> when users and moderators simply uphold the guidelines, staging a protest by demanding your account be deleted.
You're right this was a little dramatic. I think it is just annoying though when two times that I have posted about astroturfing, it has been the most upvoted comment only to get flagged. I guess this is a self-fulfilling prophecy though, as you are right that other human commenters are abound and tend to flag people complaining about astroturfing.
Anyway thanks for the reply. If your read of my account is that I'm more-often-than-not a bad actor, then I will stop commenting here. Seems like it is for the best. :)
> Hopefully it is clear from my other comments that I do try to provide value too.
I agree that you provide value in plenty of your comments, which is why we don't just want to ban/lose you.
> I have posted about astroturfing, it has been the most upvoted comment only to get flagged
People love a conspiracy theory, and on a site like HN that has many people looking at it at once, it's easy to get a large number of upvotes in a short amount of time if enough people find it exciting. We often see off-topic, titillating one-liners or ragebaity comments at the top of threads, and we always have to downweight them to keep the discussion on-topic and healthy.
> Anyway thanks for the reply. If your read of my account is that I'm more-often-than-not a bad actor, then I will stop commenting here. Seems like it is for the best. :)
It seems like you're well intentioned. You have your concerns about A.I., as many do and that's fine. You're still very welcome here. Just please try to believe that many or most of the people who are enthusiastic about A.I. are as sincere in their positivity as you are in your concern.
fyi, i flagged it because it is boring reading and against the rules.
if you suspect astroturfing, flag the comments and contact the mods.
(complaining that your complaint got flagged is also tiresome. contact the mods. "@dang" doesnt work, use the email.)
I respect you for replying here though, and yes I get that HN forum standards would suggest flagging my previous comment. But it is just sad to see a place used to be so vibrant get manipulated because of how much weight it holds for us in the industry.
And yea sure, I could go and flag all the bots and message Dang. But probably time to stop shouting into the void. :)
yes, i think so.
because, unfortunately, complaining about bots (or astroturfing, or whatever) doesn't stop them. so we end up with threads that have both the potential bot/astroturfing/whatever activity and complaints, which further drowns out any interesting comments.
Also, I don’t see that much astroturfing here? (And I tend to see it a lot on HN.)
I see long massive pro apple threads. I don't get it at all. As in; literally don't understand what Apple is good for. But I have friends, family irl who love apple so I know the sentiment exists. I therefore accept that many HN users are similar.
Many people really truly are happy to see another model drop and are excited about progress etc etc. Surely you've met such ppl in real life. Well, they're here too (I'm one of them fwiw)
This is why HN has been on the down hill in quality and those that care to highlight that are being punished, while the astro-turfing, gaslighting and Show HN self-promotion slop continues.
Otherwise, yes, we agree. Although, given the other replies to this, there are clearly those who disagree who appear to be smart and level-headed.
Anyhow I've learned my lesson now.