News for hackers. Stuff that matters.
Re:GPT-6 Sol and Luna by simonw on Tuesday September 22, 2026 @02:41PM (#49806126)
GPT-6 Luna being half the price of GPT-5.6 Luna is a really big deal.

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.

Re:GPT-6 Sol and Luna by gizmodo59 on Tuesday September 22, 2026 @02:48PM (#49806211)
6-luna is at the pareto for most of the tasks! I dont know how they make money here but its insane value from a closed source model. I'd go further and say it makes no sense (privacy, sovereignty etc aside) to use many other models as its not only expensive but also many providers don't have that much GPUs to serve at a significant volume. https://openrouter.ai/rankings?view=month#top-models 5.6 luna is already the most used model this month.
Re:GPT-6 Sol and Luna by booty on Tuesday September 22, 2026 @03:14PM (#49806631)

    I dont know how they make money here
Well, here's the neat thing: they don't!

Snark aside, Luna 5.6 was (is) an incredible game-changer.

Re:GPT-6 Sol and Luna by atoav on Tuesday September 22, 2026 @04:34PM (#49807690)
"We lose money on ever sale, but we plan to make it up in volume"
Re:GPT-6 Sol and Luna by Barbing on Tuesday September 22, 2026 @10:00PM (#49810762)
*govt bailouts
Re:GPT-6 Sol and Luna by adventured on Tuesday September 22, 2026 @10:39PM (#49810986)
They're closing in a billion users. That's Google search territory.

OpenAI is sitting on a $100+ billion ad network, incoming.

They're not going to need a government bailout, they're going to be a spigot of cash production.

Every single thread on HN keeps saying the same ridiculous thing, going on a year now. It's like they've never heard of advertising, which SV specializes in. It's like they're oblivious to the fact that every mega platform with so many users becomes an ad goldmine, and GPT's context positioning is even richer than search.

Re:GPT-6 Sol and Luna by locknitpicker on Wednesday September 23, 2026 @01:39AM (#49812064)
> OpenAI is sitting on a $100+ billion ad network, incoming.

How can you make this sort of claim with a straight face, knowing that a chinese model downloaded for free from ollama works as well if not better than OpenAI's models, without costing you a cent.

Re:GPT-6 Sol and Luna by Barbing on Wednesday September 23, 2026 @02:08AM (#49812255)
I guess it’s assuming the fact ChatGPT is a household name will bring it near permanent relevancy? I’m skeptical.

And sorry to the parent commenter if I’m making a bad assumption.

Re:GPT-6 Sol and Luna by supernovae on Friday September 25, 2026 @08:14AM (#49843554)
I have to ask you the same question.

GPT is still 20 a month for most people, 50-100 or 200 for pros.

Meanwhile, for local LLM's - everyone is chasing hardware that is crazy expensiv eand to get around it they're leasing it or putting it on credit card. DGX Sparks are insane, and you need 2 for most people talking in this thread. Mac Ultra 5 is amazing but 6k minimum 12k for the build most want so many people lease it for 240 a month which doesn't even include electricity or time in setup so instead of talking about facts, we get into weird arguments like console wars where people give APple a 5 trillion dollar company 250 a month for 36 months and don't even own their hardware money "because they can run local models" vs just paying for output from their choice of frontier.

I say this fully loving local llms and embracing them, but the reality is, local llms have gotten so expensive and continue to get expensive while we keep talking about this "Threat" of apis - where there are a lot more than openai and anthropic available much cheaper and competitive priced.

Oh, and they don't work better than OpenAI or else we wouldn't even have these discussions.

Re:GPT-6 Sol and Luna by larodi on Tuesday September 22, 2026 @04:35PM (#49807706)
> Well, here's the neat thing: they don't!

perhaps it then does mean - squeeze as much as you can get off this actual free usage.

Re:GPT-6 Sol and Luna by adventured on Tuesday September 22, 2026 @10:42PM (#49811002)
Luna is about suppressing inexpensive Chinese model competition.

It's super simple.

Gigantic hyper margin ad network = artificial subsidization of cost for various tiers = put the boot on the neck of Chinese competitors. There's no scenario where they can compete with what advertising margins make possible in terms of artificially lowering prices charged.

Re:GPT-6 Sol and Luna by locknitpicker on Wednesday September 23, 2026 @01:36AM (#49812050)
> Luna is about suppressing inexpensive Chinese model competition.

I think so too. To me the so-called Chinese local models are a clear move to prevent US companies to establish a foothold and build a moat around their business. US companies are clearly invested in a strategy to make themselves relevant with claims of major impressive achievements with the so called frontier models, and how these and only these are unblocking whole ranges of applications. At the same time, they are heavily invested in pushing AI on all absurd types of mundane tasks, such as transcribing meetings and... talking to your own kids?

In the meantime it's rather obvious that, in spite of all the propaganda, frontier models are required only in ultra niche applications, whereas the ability to run any model at all already provides most of the value. In fact, US companies have been renownee by dumbing down older generation models in what seems to be a desperate attempt to make newer models look better and influence their uptake rate.

So there is no better way to take the wind out of the US AI companies' sail than pulling a two-punch attack consisting of not inly releasing capable models that refute the "only US frontier will do the job" thesis but also releasing them for free to commodities them and eliminate the business impact of dumbing down models.

Re:GPT-6 Sol and Luna by idbnstra on Wednesday September 23, 2026 @10:22AM (#49816652)
> and... talking to your own kids?

i don't doubt they're pushing for using AI for that, but i'm curious of examples of where they're doing this. commercials, ads, etc.

Re:GPT-6 Sol and Luna by locknitpicker on Wednesday September 23, 2026 @12:33PM (#49818654)
> i don't doubt they're pushing for using AI for that, but i'm curious of examples of where they're doing this. commercials, ads, etc.

You just be living under a rock. Not do long ago Sam Altman was floating this fantastic usecases for AI was to have it explain to you your kids interests, and have it create a podcast for you to be able to keep in touch.

Re:GPT-6 Sol and Luna by mordae on Wednesday September 23, 2026 @03:12AM (#49812681)
Chinese buy their tokens at home. West as a market is an afterthought for their companies them. Western AI is banned, so only used via resellers by small fish, not companies. US has zero presence at that huge market, and absolutely not a moat.

They are buying Huawei accelerators in bulk to serve their local customers. The whole system is currently optimized to deliver a lot of cheap LLMs and hardware for them to run on.

Re:GPT-6 Sol and Luna by supernovae on Friday September 25, 2026 @08:10AM (#49843514)
Everyone keeps saying this, but I don't believe it's true. I think Luna is just an MLA or sparse architecture like all the flash variants and it's just cheap to run.

In any case, over the past few years, the only thing that has gotten more expensive is the hardware to run local models while API and Subs have gotten more affordable or feature rich while remaining same price.

They can compete because they have the compute to run the volume and if it's good on agentic work, people will be less incentivised to use other models for "Tasky-y" work.

It's literally increasing their market opportunity

Re:GPT-6 Sol and Luna by krat0sprakhar on Tuesday September 22, 2026 @02:50PM (#49806239)
Can't agree more. Between 5.6 Luna and Gemini 3.8 flash I'm so happy for the value I'm getting for my dollar (subscription pricing not API pricing) :)
Re:GPT-6 Sol and Luna by jadbox on Tuesday September 22, 2026 @03:04PM (#49806458)
Gemini 3.8 Flash looks like its better than v7 Luna/Sol on DeepSWE v1.1 while at $0.75 per million input tokens and $3.75 per million output tokens. Luna is much cheaper, but Flash has nearly Astra's performance for under the price of Sol ($2/$10).
Re:GPT-6 Sol and Luna by antupis on Tuesday September 22, 2026 @03:24PM (#49806761)
Flash thinks much more so it’s pretty much line with Sol for performance. That said I like flash coding style much more than OpenAi models.
Re:GPT-6 Sol and Luna by jeffnash on Tuesday September 22, 2026 @03:42PM (#49807032)
out of curiosity, what type of code/language do you usually use flash to write?
Re:GPT-6 Sol and Luna by spockz on Tuesday September 22, 2026 @05:27PM (#49808349)
I use it for golang, and it is fantastic. Incredibly fast. It seems the llm and I “understand” each other. I have to be less careful in my exact phrasing. It kind of just does what I want and expect.

When I ask for an explanation it adds the right amount of detail. Of course, some of the material is new to me so subtle errors are hard to spot. But at least I’ve caught Terra and Sol on inconsistent messaging.

Also I’ve found 3.8 flash to circle back to root issues even at the conceptual level like problem fit and conceptual solution direction or architecture when I wasn’t achieving my goals. It flat out said I was attempting to use the wrong tool. Whereas Sol and Astra kept rabbit holing and looking for tiny implementation errors. Even after prompting them specifically to look at it broader.

Re:GPT-6 Sol and Luna by timattrn on Tuesday September 22, 2026 @06:02PM (#49808802)
what harness or plan are you using 3.8 flash with?
Re:GPT-6 Sol and Luna by Kostchei on Wednesday September 23, 2026 @12:50AM (#49811773)
anti-gravity with gemini 3.8 or gtfo
Re:GPT-6 Sol and Luna by spockz on Wednesday September 23, 2026 @12:58AM (#49811820)
I’m using antigravity. I’m still on the AI Pro plan for the promotional $5/month.
Re:GPT-6 Sol and Luna by desterothx on Wednesday September 23, 2026 @02:43AM (#49812509)
where is this promotion?
Re:GPT-6 Sol and Luna by spockz on Wednesday September 23, 2026 @02:53AM (#49812568)
If you don’t have a plan yet, log in to antigravity. There will be a button “upgrade plan” somewhere. Sometimes it pops up and otherwise lookup in settings > account. There should be some button that says upgrade. Clicking that brought me to the google studio ai page which offered the 20-something plan for €5/month.
Re:GPT-6 Sol and Luna by oh_no on Tuesday September 22, 2026 @04:01PM (#49807280)
look at token use, 3.8 flash is a huge token hog compared to openai models
Re:GPT-6 Sol and Luna by Citizen_Lame on Tuesday September 22, 2026 @03:42PM (#49807034)
Gemini 3.8 Flash and 3.1 Pro are pure rubbish. Very little thinking, mediocre and usually incorrect results. They cannot be compared to frontier models.
Re:GPT-6 Sol and Luna by anukin on Tuesday September 22, 2026 @04:45PM (#49807829)
This is my experience as well. I am surprised that lot of people find it much better than Luna.
Re:GPT-6 Sol and Luna by mapontosevenths on Tuesday September 22, 2026 @07:13PM (#49809585)
I suspect that the people saying this haven't used Luna.

It's also weird that anyone uses it outside of an enterprise. They force you to use Googles inferior harness on the plans and I doubt any mere mortal is paying that much, for so little usage, with the worst harness on the market.

Re:GPT-6 Sol and Luna by bonestamp2 on Wednesday September 23, 2026 @03:07AM (#49812646)
I prefer 5.6 Luna while a coworker prefers 3.8 Flash. The difference seems to be that they chat with Flash (with code context) while I just ask Luna to directly modify the code. I was already very impressed with 5.6 Luna so I am looking forward to running 6.0 Luna all day tomorrow to see how it compares.
Re:GPT-6 Sol and Luna by krat0sprakhar on Tuesday September 22, 2026 @04:57PM (#49807975)
TBH: I really like how fast 3.8 Flash is... Once I have clear plan, I feel quite confident in delegating large parts of implementation to Flash and Luna
Re:GPT-6 Sol and Luna by mgkimsal on Tuesday September 22, 2026 @08:32PM (#49810118)
Maddening for a bit - I've got problems that Flash is better on, and some Luna is better on, but I generally don't know until one has wasted time/tokens. Then I switch to the other one and... it's often just... bam - done. Correctly. I can't find the patterns ahead of time to determine what model I should be using first. :/ That said, I've been alternating between both the last month or so and they've both been pretty good compared to earlier models.
Re:GPT-6 Sol and Luna by Kostchei on Wednesday September 23, 2026 @12:51AM (#49811781)
codex seems pretty solid on review, flash is fast on basics but makes more mistakes/errors, Claude is just to picky for me
Re:GPT-6 Sol and Luna by zozbot234 on Tuesday September 22, 2026 @03:54PM (#49807190)
MiMo 2.6 Pro is at the Pareto frontier (the one where you only need 20% of the smarts for 80% of the tasks) according to Artificial Analysis, nicely filling in as a substitute for a hypothetical 'GPT-6 Terra' (which doesn't exist as far as we know). That's pretty darn impressive from an open model.
Re:GPT-6 Sol and Luna by Ternari on Tuesday September 22, 2026 @04:38PM (#49807735)
That's not what the Pareto frontier is; you're mixing up Pareto frontier with Pareto principle.

https://en.wikipedia.org/wiki/Pareto_front

https://en.wikipedia.org/wiki/Pareto_principle

Re:GPT-6 Sol and Luna by Rexxar on Tuesday September 22, 2026 @05:02PM (#49808058)
Despite the error in the parenthesis, it's exactly what he says: https://artificialanalysis.ai/?intelligence-category=text-on...
Re:GPT-6 Sol and Luna by Ternari on Tuesday September 22, 2026 @05:15PM (#49808218)
I was just responding to the error in the parenthesis.
Re:GPT-6 Sol and Luna by supernovae on Friday September 25, 2026 @08:17AM (#49843589)
mimo 2.6 is kinda dumb and over tuned though but not bad for a checkpoint testing their RL
Re:GPT-6 Sol and Luna by arcanemachiner on Tuesday September 22, 2026 @04:08PM (#49807371)
> I dont know how they make money here

I assume it's a subsidy to get more training data.

Re:GPT-6 Sol and Luna by tedsanders on Tuesday September 22, 2026 @07:57PM (#49809878)
By default, OpenAI does not train on API data. I promise you that Luna's low pricing is not a subsidy to get more training data. We've been lowering prices for years.

(I work at OpenAI.)

Re:GPT-6 Sol and Luna by arcanemachiner on Wednesday September 23, 2026 @12:20AM (#49811598)
Wait, so you guys don't anonymize the user data, then train on it after it's been sanitized? I thought this was done to some degree or another.

So what is the value prop then? Just basic supply and demand?

FWIW I have definitely noticed OpenAI's emphasis on efficiency and value in the last year, so that part isn't new to me... I just thought there was more to it then that.

Re:GPT-6 Sol and Luna by tedsanders on Wednesday September 23, 2026 @03:11AM (#49812672)
API: By default, no training (opt in).

ChatGPT enterprise: By default, no training (opt in).

ChatGPT personal: By default, training (opt out).

Re:GPT-6 Sol and Luna by shostack on Wednesday September 23, 2026 @11:16PM (#49825797)
Ted can you confirm your choice of words here to be precise for an audience who is familiar with the nuances, when you say "no training" or "training (opt out)" for personal... Is that inclusive of "sanitized" (or pseudonymized) data?

Your response to the original question is using generalized terminology when there is a very important distinction the OP made by the use of "sanitized."

People want to know to that extent derivatives of their data are being used. Synthetic data has been proven to be effective at generating training data and AI is very good at shuffling context such that you have something where you don't have to say it is "user data."

But there are many shades of gray there for people versed in how the sausage is made. I'm sure you'll appreciate then why your response leaves additional questions in light of that "sanitized" distinction.

Re:GPT-6 Sol and Luna by tedsanders on Thursday September 24, 2026 @05:46PM (#49837137)
Yeah, I'm not trying to trick anyone with wording that's technically true but actually misleading.

When I say no training, I mean no training. No gimmicks around data vs derived data, synthetic data, preference data, etc.

Places where I can imagine potential cracks in the literal interpretation of what I said are things like a financial analyst who does a statistical fit to predict revenue next quarter using a model based on last quarter's aggregate token consumption, which in some sense embodies your metadata (the length of your conversations) in a sea of other data. Or perhaps an infrastructure planner who makes a little model of internet bandwidth by time of day to help plan when we need a data center networking upgrade. Maybe things like these are technically training on your data in the most pedantic sense, but definitely not in the sense that most of us mean.

I promise you we're not doing any gimmicks where we transform your data and then pretend ah because it's transformed it's not your data.

Re:GPT-6 Sol and Luna by shostack on Thursday September 24, 2026 @08:59PM (#49838915)
TY, appreciate the thoughtful reply. Transparently, I have been so on the fence with moving more of my workflows and personal usage over from Hermes+VPS+ZDR model provider, or implementing the "Hermes shell over Codex Subscription" pattern because I worry about:

1. Legal loopholes given OpenAI's advertising aspirations and model training needs

2. Data retention and rising threats of fascism that historically have not served the persecuted very well when fascist regimes get access to said data

3. Risk from centralized collection of that data with a company whose software I do not control in a world where enshitification and lock-in is the norm.

I really wish OpenAI did more to espouse exactly this: "When I say no training, I mean no training. No gimmicks around data vs derived data, synthetic data, preference data, etc." and ideally provide technical reasurrances that this is impossible (eg: certain technical ZDR approaches, etc.).

Do you happen to have a favorite reference to point me at that would document some of those official assurances to the nuanced detail we've discussed here?

Re:GPT-6 Sol and Luna by tedsanders on Friday September 25, 2026 @02:54PM (#49848477)
API: https://developers.openai.com/api/docs/guides/your-data

ChatGPT: https://help.openai.com/en/articles/5722486-how-your-data-is...

ChatGPT data controls: https://help.openai.com/en/articles/7730893-data-controls-in...

If you have feedback on how to improve these, happy to consider it.

Looks like we phrase it as "your new conversations won’t be used to train OpenAI models" which is hopefully clearer than "OpenAI models will not be trained on your conversations", which could leave open the possibility of derived data or something.

Re:GPT-6 Sol and Luna by InsideOutSanta on Tuesday September 22, 2026 @03:17PM (#49806677)
> I dont know how they make money here

By raising it from investors.

Re:GPT-6 Sol and Luna by GolfPopper on Tuesday September 22, 2026 @05:22PM (#49808293)
To whom they promise the Sun, the Moon, and the Stars. Roflmao. Whatever the merits of the underlying technology, the business model is pure hucksterism.
Re:GPT-6 Sol and Luna by m101 on Tuesday September 22, 2026 @03:42PM (#49807023)
perhaps they use this as the carrot to get you locked into their monthly plan over anthropic's.
Re:GPT-6 Sol and Luna by abirch on Tuesday September 22, 2026 @03:43PM (#49807043)
works great until they raise prices.
Re:GPT-6 Sol and Luna by usef- on Tuesday September 22, 2026 @05:52PM (#49808680)
There's no difficulty in cancelling.
Re:GPT-6 Sol and Luna by trollbridge on Wednesday September 23, 2026 @07:14AM (#49814309)
First they’d have to let you sign up for a new 20x Pro account.
Re:GPT-6 Sol and Luna by 7777777phil on Tuesday September 22, 2026 @03:53PM (#49807170)
I guess I have to update my pareto front then: https://philippdubach.com/posts/jev-model-router-for-pi/
Re:GPT-6 Sol and Luna by the__alchemist on Tuesday September 22, 2026 @04:26PM (#49807576)
How does 6-Luna xhigh compare to 6-Sol medium? Or more broadly newer/bigger model with lower effort vs older/smaller higher effort?
Re:GPT-6 Sol and Luna by knicholes on Tuesday September 22, 2026 @05:36PM (#49808493)
Read the link! It's in there.
Re:GPT-6 Sol and Luna by lacker on Tuesday September 22, 2026 @04:39PM (#49807755)
Offering Luna for cheap is like restaurants giving you free bread and water. They're pretty sure that you're going to end up eating the expensive stuff on the menu.
Re:GPT-6 Sol and Luna by usef- on Tuesday September 22, 2026 @05:51PM (#49808665)
Note that to sit at a restaurant you're obliged to order something, though. Here there is no obligation to go beyond the model you choose.
Re:GPT-6 Sol and Luna by sieve on Tuesday September 22, 2026 @05:31PM (#49808409)
My OpenCode Go stats for the last 30d:

Cached Read: ~6,500M

Input: ~150M

Output: ~20M

Approx $40 worth of usage across DeepSeek V4 Flash + MuseSpark Contributor 1.3. And a bit of both the GLM models. This is covered in a $10 subscription.

If I were to use Luna's API pricing:

$0.02 x 6,500 = $130

$0.20 x 150 = $30

$1.20 x 20 = $24

So $184. And this is assuming smaller coding sessions (<272K) beyond which Luna pricing doubles.

--

Cost wise, these models are nice for small stuff. Translations etc. Any model that does not provide multiple Mtoks of cached reads per cent is not very useful to me for coding workflows.

Re:GPT-6 Sol and Luna by MitziMoto on Tuesday September 22, 2026 @07:51PM (#49809840)
These are also the orders of magnitude of our production agents for our business (NOT coding). Cache reads are so heavy compared to anything else that it's the only price point that really matters, regular input and output are negligible.

I need aggressive cache read pricing with full prompt_cache_key support to have a model be financially viable for our workload. Right now Meta Muse 1.3 Contributor is the only one that makes sense--but we are starting Evals on the new MiMo 2.6 class to see how it holds up.

Re:GPT-6 Sol and Luna by sieve on Tuesday September 22, 2026 @08:20PM (#49810038)
I have used MiMo 2.5 extensively. MuseSpark and DS4 Flash are MUCH smarter than that one. But MiMo follows instructions diligently. So it has been useful as the implementer of a spec designed by Claude/Kimi.

One good thing about MiMo that I experience on OpenCode is the provider seems to cache tokens for much longer than MS13/DS4F. I have seen cache being hit for close to an hour after the last request. The corresponding timing for MS13/DS4F is in the 1-5 min range.

I am trying out MiMo 2.6 Flash as well.

Re:GPT-6 Sol and Luna by gleenn on Tuesday September 22, 2026 @09:02PM (#49810359)
Last I heard, caches had like a 5 minute TTL... doesn't that mean if you get up and make a coffee (hand pour over of course), that you are back at full price?
Re:GPT-6 Sol and Luna by jmalicki on Tuesday September 22, 2026 @09:13PM (#49810443)
I wish that was more programmable.

You can pay for higher cache time, you can pay for NVMe KV cache for an hour that can just be reloaded, etc., at a lesser tier you can pay for the KV cache to be stored on a network store (I guess I'm unclear if that last tier would be cheaper than recomputation, not even 100% sure of the NVMe with direct GPU<->storage DMA) depending on your model settings.

Re:GPT-6 Sol and Luna by gizmodo59 on Tuesday September 22, 2026 @07:03PM (#49809507)
It’s not direct token to token pricing and everyone misses it. The cost is how much tokens to complete something multiplied by token pricing. I can have a model at .0001 per million tokens but it’s so inefficient that it takes 10B tokens to complete a task means it’s expensive.
Re:GPT-6 Sol and Luna by sieve on Tuesday September 22, 2026 @07:40PM (#49809771)
I am not designing rockets. Most of my work is bog standard hobbyist stuff: compilers, vms, sandboxes, system tools of various kinds, SSGs, markup languages, plain text ledgers etc. Even Gemma/Qwen running locally can manage this.

Frankly, I have no idea what people do with Opus/Fable etc. I don't think anything I do needs something that charges $50/M for output tokens.

Re:GPT-6 Sol and Luna by apatheticonion on Tuesday September 22, 2026 @09:51PM (#49810699)
Can confirm. I have been using DeepSeek since forever and it's so good I was able to write a compiler and native desktop applications with it. I use it as a coding assistant in my IDE so the results end up at the same quality I would write by hand.

I recently started a job that only uses Claude models. Opus and Sonnet are so slow you have no choice but to do multiple tasks in parallel. You create a git worktree, set off an agent to do something, another worktree, set out an agent - then play video games for 20 minutes until they complete the task (poorly).

You can't really do "guide coding" like you can with DeepSeek-style flash models because Claude is too slow.

I think the idea with slow frontier models is to end up with "software factories", where you just write tickets and send them to a harness that delegates work to agents/subagents. Your job is to prompt and review (and eventually just prompt).

Mathematically and assuming token prices/efficiency remains constant, the collective US AI industry needs to increase token usage by 15x before 2030 (3.5 years from now) to satisfy investors. With companies already implementing token limits, the only place from here is for frontier models to replace staff entirely to expand budgets for tokens. The only way to do that is to demonstrate the efficacy of software factories and headless agentic workflows.

Objectively, I have set up a software factory and I do see the utility of it, though I did it with DeepSeek and prices are 1% that of frontier models - which doesn't bode well for investors looking for an eventual return.

Heck, my old M1 MBP 32gb running Qwen 3.6 35b a3b sipping 10w when generating tokens is good enough for a lot of my guide-coding work - it's just a bit slow so I use DeepSeek instead. When hardware prices come down, I honestly wouldn't see a need to subscribe to any service, I'd just grow my own tokens at home.

Re:GPT-6 Sol and Luna by sieve on Tuesday September 22, 2026 @10:00PM (#49810759)
I use Claude Sonnet and ChatGPT via the web UI. I often use Claude to come up with specs for my ideas. This is becoming less and less useful. DS4/MS13/MiMo are almost there for these use cases as well.

I dogfood everything I produce, and the models are good at collaborating with me on a spec and then turning it into code.

If Sonnet/ChatGPT suddenly became unavailable due to Anthropic/OpenAI suddenly not being able to subsidize the freemium/loss-leader experience, I probably would not miss them. Google/BraveAI already give you the AI experience during search (when you are looking for stuff to buy, or something particular). Claude/ChatGPT still have a minor edge in this use case for me right now.

Re:GPT-6 Sol and Luna by jklmnopqrstuvw on Tuesday September 22, 2026 @11:09PM (#49811180)
From my own testing, Claude/GPT is much faster than Deepseek.
Re:GPT-6 Sol and Luna by apatheticonion on Sunday September 27, 2026 @05:41PM (#49871143)
I haven't tried GPT Luna yet, but anything Claude takes forever in my experience.

In guided coding sessions, DeepSeek is so fast I rarely have the time to look away from my screen. I've noticed DS 4.1 is a little slower, but claude is still orders of magnitude slower.

e.g.

- "Turn this SQL CREATE TABLE statement into a repository class and generate models for it"

- "Add error handling / timeout to this function"

- Highlight text "Using this as a reference, repeat for all other files in this folder"

Re:GPT-6 Sol and Luna by tripzilch on Wednesday September 23, 2026 @09:32AM (#49815895)
> running Qwen 3.6 35b a3b sipping 10w when generating tokens is good enough for a lot of my guide-coding work

can you tell more about how you're using it? like, what harness? or also in the IDE?

I found Qwen3.6 35B/A3B to make slightly too many mistakes (already in its harness' tool use, hence my question), maybe it gets the job done, but it will also sometimes generate a bit of a mess (e.g. editing/creating files in the wrong folders) and fixing/solving its own mistakes takes time (or tokens) ..

Re:GPT-6 Sol and Luna by jorgeleo on Wednesday September 23, 2026 @10:45AM (#49816984)
Same thing for me. on an M5 max, Qwen 3.6 35b gives me between 150 and 200 tps using splash as inference engine.

More than enough for guided code sessions, at 100% privacy. And i can use obliverated models if i am trying to harden my own app, something i cannot do with cloud providers.

Re:GPT-6 Sol and Luna by Phemist on Wednesday September 23, 2026 @04:34PM (#49822177)
Cool did not know about Splash. Seems interesting!

https://github.com/incoai/splash/issues/38

Looks like an issue exists to convert model weights for ornith1.5 as this is a magical process atm.

Re:GPT-6 Sol and Luna by Phemist on Wednesday September 23, 2026 @04:28PM (#49822067)
I feel like ornith1.5 35B/A3B is an overall stronger model on the same architecture, so a drop-in replacement untill qwen3.8/qwen4 is released. Using the 8bit quant on my M4 max gets around 80tok/sec output/decode on an empty context, dropping down to 35ish on nearly full one.
Re:GPT-6 Sol and Luna by apatheticonion on Sunday September 27, 2026 @05:49PM (#49871211)
Guide coding is more forgiving because the diffs are small enough that you can just try something and veto it if it's not good. Qwen 3.6 a3b makes more mistakes than DeepSeek but it's free. I'd imagine the next ~32gb-vram-class MoE model from Qwen will close the gap.

The real deciding factor for me is inference speed.

I use VSCode insiders with their BYO model configuration. I stage little bits at a time in a tight prompt-review-prompt-review workflow. I occasionally use dedicated harnesses (DeepSeek harness, Codex, etc) because they have better tools and outcomes than VSCode's built-in harness for longer horizon tasks - though I lack the ability to highlight a block of text and say "add error handling" or similar.

When building visual applications, desktop harnesses are better because it's a bit easier to send screenshots to the agent.

I love Zed editor but its AI review features are lacking compared to VSCode. I go back to it frequently and am ready to switch over when the team resolves the usability issues.

Re:GPT-6 Sol and Luna by unsupp0rted on Tuesday September 22, 2026 @11:21PM (#49811254)
It sounds like you’re still writing code by hand and reading and reviewing code.

For that any decent model from the past year will do.

If you want to forget how to write code and not read generated code, then you need a very good frontier model, ideally one from 6-12 months in the future.

Re:GPT-6 Sol and Luna by sieve on Wednesday September 23, 2026 @12:45AM (#49811750)
I used to read the code till around May. Now I don't. Instead I validate behavior. And have multiple LLMs verify that the code implements my handwritten spec.

MiMo 2.5/2.6, MuseSpark 1.3, DeepSeek V4/4.1 Flash and GLM 5.3 Flash are perfectly capable of following my spec and then poking holes in the implementation till there are none left.

Re:GPT-6 Sol and Luna by cluckindan on Wednesday September 23, 2026 @08:08AM (#49814855)
Thanks for creating work for actual engineers.
Re:GPT-6 Sol and Luna by locknitpicker on Wednesday September 23, 2026 @12:59AM (#49811833)
> It sounds like you’re still writing code by hand and reading and reviewing code.

This is such a naive, baseless opinion.

Nowadays any AI coding assistant service supports or can be used with sub-agent orchestration frameworks.

If you are in the business of software factories, you can use the cheapest models and even local models to handle some if not all tasks in the orchestration chain.

Adding tests or executing tests (unit, integration, UI, you name it) doesn't require a cutting edge frontier model. Neither does refactoring. Neither does identifying call stacks. Neither does planning a changeset.

You have your specialized subagents, you put together a small orchestrator subagent that handles feedback loops and handoffs,and you throw it at tasks.

For the past couple of months, most of the code I write is not code per se, it's subtask orchestrators. And unlike the old "only Opus is passable" days, the cheapest models do get the job done.

Re:GPT-6 Sol and Luna by darkwater on Wednesday September 23, 2026 @02:50AM (#49812545)
This kind of opinion has been around for about 10 months now already, since Opus 4.5 and Claude Code initial release. It just shifts alongside models.
Re:GPT-6 Sol and Luna by dcl on Tuesday September 22, 2026 @07:54PM (#49809861)
How have you found Muse Spark 1.3? It doesn't get much mention, despite pretty good benchmarks. I've been using a bit at home and find it quite good, often finding mistakes made by Opus 5.
Re:GPT-6 Sol and Luna by sieve on Tuesday September 22, 2026 @08:14PM (#49810000)
MS13 is pretty sharp and has been my workhorse for the past month. It follows my coding style and commit/clean workflows referenced in AGENTS.md perfectly but has the habit of doing things without conferring with me (the Gemini problem). So you need some kind of instruction for that.

It starts failing around the 5-600K context mark, but you can have it generate a handover document and continue in the next session.

I would not use it at sticker price, but the Contributor version is priced just about right.

Re:GPT-6 Sol and Luna by slopinthebag on Wednesday September 23, 2026 @01:49AM (#49812130)
shocking. the code it generated, while technically working, was entirely garbage. i used it for code review and it flagged twenty issues, sol checked the review and found 75% of them were hallucinations. sol was much closer to reality. i no longer trust benchmarks at all because of it.
Re:GPT-6 Sol and Luna by dcl on Friday September 25, 2026 @08:56AM (#49843987)
Interesting. I have been asking it to review code from Opus 5 and it found tonnes of issues, Opus agreed with the findings too.
Re:GPT-6 Sol and Luna by handfuloflight on Tuesday September 22, 2026 @06:08PM (#49808879)
How long can OpenCode bleed for?
Re:GPT-6 Sol and Luna by sieve on Tuesday September 22, 2026 @06:24PM (#49809089)
Are they bleeding? Their multipliers seem to be reasonable. They are not offering $60 worth of usage for $10 on every model, only some. In the case of the expensive ones, it is only $15.

Given how subscription models work (not every one uses every last $ of their plan), they should achieve breakeven soon enough I guess.

Re:GPT-6 Sol and Luna by ronsor on Tuesday September 22, 2026 @06:33PM (#49809211)
They already stopped. That's why the service quality declined.
Re:GPT-6 Sol and Luna by handfuloflight on Tuesday September 22, 2026 @06:45PM (#49809336)
What did you notice?
Re:GPT-6 Sol and Luna by nearbuy on Tuesday September 22, 2026 @08:44PM (#49810217)
This isn't right. You're comparing cost per token, but DeepSeek V4 Flash uses more tokens. Artificial Analysis found GPT 6 Luna to be significantly cheaper than DeepSeek: https://artificialanalysis.ai/models/comparisons?compare=dee...
Re:GPT-6 Sol and Luna by sieve on Tuesday September 22, 2026 @09:43PM (#49810637)
I do not (generally) trust benchmarks. I only trust what a model does with MY code.

Forget DS. I asked MiMo 2.6 yesterday to explain ML/LLMs to me succinctly and the pointed it at Karpathy's micrograd code. It produced a C implementation called `xor_mlp`, a tiny model that learnt how `xor` worked. I then asked it to produce a model that can play tictactoe without losing (mostly). It did. It supervised the training process and produced a compiled version with multiple switches. The pi-dev session is still running, so here are actual stats

↑45k ↓35k R1.0M CH99.4% $0.019 4.2%/1.0M (auto) - (opencode-go) mimo-v2.6-flash • high

And here is Luna on the same workflow (I had to poke and prod a bit to get what I wanted):

↑141 ↓34k R1.0M W43k CH95.3% $0.072 4.2%/1.1M (auto) (opencode-go) gpt-5.6-luna • high

I expect similar results from DS41F/MS13. Closer to MiMo costs than Luna.

So the "significantly cheaper" thing may not really hold, more so when Luna has to actually read my codebase to do the stuff that I want rather than rely on world knowledge. The 8-10x cache read cost differential itself will kill the token budget.

Re:GPT-6 Sol and Luna by asaddhamani on Wednesday September 23, 2026 @02:56AM (#49812584)
Don’t know if it’s still true but with Chinese models, using Western API providers is significantly more expensive and using Chinese providers they will train on your inputs without exception. That has kept me from using these ultra cheap endpoints.
Re:GPT-6 Sol and Luna by sieve on Wednesday September 23, 2026 @03:07AM (#49812651)
Do you really think Western providers will not train on your data? I have no such illusions.

I try to keep PII out of what I share with LLMs. Otherwise, I do not see the point, really. Very little of my code is "unique." I simply approach things a bit differently. Otherwise the algorithms and code would be similar to what others with domain knowledge would write. So much of code and algorithm implementations are available in the open. And LLMs have trained on all of them.

What they most probably gain from you is your prompts and your thinking approach more than the code.

Re:GPT-6 Sol and Luna by intended on Wednesday September 23, 2026 @04:44AM (#49813267)
The meager difference is that, in theory, you can eventually sue people in the US.

In theory.

Also, this is a feature for people who live in America, and mostly irrelevant for everyone in the global south.

Re:GPT-6 Sol and Luna by gf000 on Wednesday September 23, 2026 @05:05AM (#49813398)
As a European, I honestly don't see a difference between the US and China from this perspective. They are both equally untrustworthy in my book.
Re:GPT-6 Sol and Luna by rrr_oh_man on Wednesday September 23, 2026 @08:10AM (#49814883)
As a Western European, I see the same untrustworthiness in Europe.

We just have our personal privacy security theater in the form of GDPR and a feeling of moral supremacy that's been drilled into our heads from primary school on.

Re:GPT-6 Sol and Luna by bigfudge on Wednesday September 23, 2026 @02:51PM (#49820709)
GDPR isn’t theatre in many organisations. Yes large tech firms (mostly US) probably ignore or circumvent. But most businesses I’ve worked for have taken concrete steps to reduce the data they hold and consider how it’s being used asa direct consequence of gdpr.
Re:GPT-6 Sol and Luna by lejalv on Wednesday September 23, 2026 @07:42AM (#49814579)
Don't understand why you are downvoted.
Re:GPT-6 Sol and Luna by asaddhamani on Wednesday September 23, 2026 @04:55AM (#49813341)
People use LLMs for far more personal tasks than just writing code. There are AI journaling apps for instance. And yeah, western providers give you a toggle but I don’t know if that toggle actually does anything or not. They were fine with collecting training data in many morally questionable ways before, no reason for them to stop when you’re literally handing it over to them.
Re:GPT-6 Sol and Luna by ascorbic on Wednesday September 23, 2026 @05:15AM (#49813447)
The US labs would lose billions in enterprise contracts if they were found to be secretly training on data when opted-out. It's not worth it.
Re:GPT-6 Sol and Luna by dhx on Wednesday September 23, 2026 @07:31AM (#49814451)
Great in theory, but what are US enterprises going to do _if_ their private data is later found to be used for training?

1. Not use AI technology and fall behind the rest of the world.

2. Use Chinese AI technology, either hosted by Chinese companies or the models self-hosted.

3. Sue US AI companies for damages, but not enough to have any meaningful impact to such companies that it'd impact US national security goals (per US government contribution to NY Times copyright lawsuit).

Re:GPT-6 Sol and Luna by Kyo91 on Wednesday September 23, 2026 @09:34AM (#49815924)
There's a huge difference between AI companies exploiting a grey area like training on public corpora and violating a private contract that they explicitly entered into with another party. The latter is very explicitly illegal and would never survive trial in Delaware Chancery court. And all of that is before we get into Federal contracts where training on TS/SCI data could lead to criminal charges.

There's a huge market in the US for providing AI services while respecting client privacy. It makes sense for at least one major provider to offer this.

Re:GPT-6 Sol and Luna by heon29 on Wednesday September 23, 2026 @09:58AM (#49816294)
> Great in theory, but what are US enterprises going to do _if_ their private data is later found to be used for training?

This. And it’s already happening:

> 2. Use Chinese AI technology, either hosted by Chinese companies or the models self-hosted.

Re:GPT-6 Sol and Luna by tomnipotent on Wednesday September 23, 2026 @01:57PM (#49820002)
> Sue US AI companies for damages

Yes that's the whole point, at least it's an option in the US and Europe. Good luck getting any redress from China. Anthropic was already hit with a $1.5B class-action which would be impossible against a Chinese business.

Re:GPT-6 Sol and Luna by vikramkr on Wednesday September 23, 2026 @04:08PM (#49821780)
3. They would sue. And it could have very meaningful impact. NYTimes copyright lawsuit is not a valid comparison because because that's a violation of national/state law which really only matters to the extent that the government is enforcing that stuff which is not the biggest concern rn (these companies are large enough that the threat of the legal costs of fighting in court is not that scary and you'd need a government actually willing to punish them substantially for them to be scared). This stuff would be under contract law against other mega corporations with big legal teams who are also their customers which is a much scarier prospect imo
Re:GPT-6 Sol and Luna by dhx on Thursday September 24, 2026 @01:44AM (#49826684)
I somewhat agree, but with limitations:

1. Possible use of differential privacy[1] techniques to train on private data but prevent the release of statistically underpresented facts/data/words. For example, ACME Inc's private data could frequently include the term 'ACMEwidgetPRO' for an upcoming product that is not publicly revealed anywhere else. It would therefore be a bad day for the AI technology company to output 'ACMEwidgetPRO' from one of their public models. Consider now that a few models could be trained--X for public data only, Y for public and private data of ACME Inc together, Z for private data of ACME Inc. A prompt is provided to model Y but output is cross-checked with model X to double check terms such as 'ACMEwidgetPRO' are known in public. If not--provide a "I don't know" response for the prompt.

2. Possible attempted defences similar to "Oops, our model was fine tuned against a model supplied by Temporary18271 Inc (company that no longer exists) and perhaps their model might have been trained on a non-public document which was accidentally exposed to the Internet" that _might_ work occasionally to fob off concern.

3. What recourse does a small or medium company or government especially in a developing country realistically have? They perhaps can't host their own LLMs locally due to availability and cost, can't individually negotiate their own favourable terms with an AI technology company (who cares that much about a potential customer with $100k budget that has no other options anyway), and perhaps can't remain competitive in their industry without heavy use of LLMs.

[1] https://en.wikipedia.org/wiki/Differential_privacy

Re:GPT-6 Sol and Luna by vikramkr on Thursday September 24, 2026 @05:42PM (#49837091)
Yes - but those aren't limitations beyond what I was getting at that's all part of the package of the bland dystopia of late 2026. I think 1 is just a case where it comes down to who has the better lawyers, as is 2. And for 3, yes, also a large government does not have much recourse if they have decided to not flex their muscles. Pretty much the only threat I see as actually viable/scary in this world environment is along the lines of megacorp v megacorp or megacorp v broligarch - and everyone else is just caught in the cross hairs/benefits by accident at best. It would be difficult to argue that the current environment is conducive to consumer protections or equal justice under law.
Re:GPT-6 Sol and Luna by atmosx on Wednesday September 23, 2026 @07:37AM (#49814529)
Oh. Yeah of course… and the US population will revolt if the figure the NSA is spying on them.
Re:GPT-6 Sol and Luna by astrange on Thursday September 24, 2026 @12:44AM (#49826349)
> Do you really think Western providers will not train on your data? I have no such illusions.

Noone wants to "train on your data". You can't learn the answers to questions by pretraining on the questions, and nobody wants to teach the models to output text that looks like a user query.

The Chinese providers "train on your data" by sending your query to Anthropic and training on the answers that come back.

Re:GPT-6 Sol and Luna by asp_hornet on Wednesday September 23, 2026 @04:29AM (#49813168)
> without exception

Is this based on something or just because “they’re Chinese and they’ll do anything to win”.

Re:GPT-6 Sol and Luna by asaddhamani on Wednesday September 23, 2026 @04:52AM (#49813324)
This is based on my last check of alibaba and Deepseek TOS. If the Chinese will do anything to win, so will the Americans. I’m not American or Chinese and I have no reason to trust either side. I do think Chinese models are better price performance and actually open which is in many cases better.
Re:GPT-6 Sol and Luna by r_lee on Wednesday September 23, 2026 @07:48AM (#49814647)
it's very well known Deepseek does it, their whole discounted pricing was seemingly priced on that.

you can't even use Alibaba on Openrouter if you enforce ZDR

Re:GPT-6 Sol and Luna by someguynamedq on Wednesday September 23, 2026 @05:23AM (#49813496)
Why on gods green earth would they not if they can?
Re:GPT-6 Sol and Luna by _3u10 on Thursday September 24, 2026 @10:32AM (#49831186)
Check the weights for your training data. You can trust but verify with many Chinese models.

Unfortunately you just have to take the “our AI is going to take your job, then kill you, and we instruct it to hack your infra” people that they aren’t training on your data anyway.

If they are hacking hugging face and Australia to scrape data trust me they have “hacked” their own systems and are training on it.

Re:GPT-6 Sol and Luna by nearbuy on Wednesday September 23, 2026 @03:58AM (#49812967)
With GPT-6 Luna (which is what the parent comment was talking about), that would come to 3.2¢, assuming GPT-6 used the same number of tokens.

I don't think you can guess more precisely than an order of magnitude from trying each once on one task.

Re:GPT-6 Sol and Luna by sieve on Wednesday September 23, 2026 @05:21AM (#49813486)
DS is VERY talkative. Luna is less so. Still do not think, based on this little experiment, that Luna could beat DS in price: API-to-API. As part of a Plus/Pro plan? Sure.
Re:GPT-6 Sol and Luna by ducktoysleftout on Wednesday September 23, 2026 @01:57PM (#49819995)
Its style of writing is part of the fun. Seeing reasoning traces fly by that each start with “Hmm…” is pretty amusing in my opinion especially if you try to vocalize it in your mind.
Re:GPT-6 Sol and Luna by Thanemate on Wednesday September 23, 2026 @02:57PM (#49820807)
In a discussion about cost effectiveness, how subjectively fun the writing feels like to the reader isn't a factor, except maybe if we were working on writing comedy.
Re:GPT-6 Sol and Luna by yunohn on Wednesday September 23, 2026 @05:23PM (#49822798)
Well, DS shows the thinking stream so it feels that way, but I’ve realized that OpenAI hiding it just gives a false impression - the non thinking output is also very wordy for OpenAI models.
Re:GPT-6 Sol and Luna by dudisubekti on Tuesday September 22, 2026 @10:30PM (#49810923)
Artificialanalysis benchmark is a combination of a several benchmarks which might or might not represent realistic coding:

"Artificial Analysis Intelligence Index combines performance across 10 evaluations: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, and AA-LCR v1.1."

Not saying it doesnt have any value but it's probably irrelevant if you use these AIs for a specific use case. Like for example Humanity Last Exam tests general knowledge, which is not very useful for coding.

It's best to go to the specific coding benchmarks and compare there.

Re:GPT-6 Sol and Luna by aucisson_masque on Wednesday September 23, 2026 @02:27AM (#49812406)
There is too many money involved, benchmark can’t be trusted.
Re:GPT-6 Sol and Luna by dudisubekti on Wednesday September 23, 2026 @07:54AM (#49814718)
I had a favorite benchmark, SWE-rebench, but sadly it's no longer maintained.

But yeah, I'll just take these benchmarks with a grain of salt. Only hands-on experience matters in the end, and these days it's very easy to switch models.

Re:GPT-6 Sol and Luna by csomar on Tuesday September 22, 2026 @11:36PM (#49811351)
Does it use less tokens or we just get no accounting of the thinking tokens in OpenAI/Claude models?
Re:GPT-6 Sol and Luna by KoolKat23 on Wednesday September 23, 2026 @03:47AM (#49812901)
According my model,

My cost is (I use nous as provider)

DeepSeek v4-flash-0731 • Your cost: $0.56

DeepSeek v4.1-flash • Your cost: $1.22

GPT-6 Luna • Your cost: $4.22

My usage is heavy on the cache. Apparently v4.1 flash uses 1.75 times as many tokens so still cheaper.

Re:GPT-6 Sol and Luna by infecto on Tuesday September 22, 2026 @09:47PM (#49810667)
How can you compare a subscription which is most likely being subsidized with consumption pricing?
Re:GPT-6 Sol and Luna by sieve on Tuesday September 22, 2026 @09:53PM (#49810715)
I gave you the $40 option. Which is what it would cost if you used APIs on OpenRouter or elsewhere. Still beats Luna by 4-4.5x
Re:GPT-6 Sol and Luna by infecto on Wednesday September 23, 2026 @10:04AM (#49816363)
Ok great. I still don’t see how subscription costs can be compared to API.
Re:GPT-6 Sol and Luna by attentive on Wednesday September 23, 2026 @12:54AM (#49811799)
apples and API pricings
Re:GPT-6 Sol and Luna by sieve on Wednesday September 23, 2026 @02:43AM (#49812508)
You can use the models I mentioned directly from DeepSeek, Meta and Xiaomi and not exceed $40. Were it not for GLM 5.3 blowing up a quarter of my monthly budget in 5h, we are actually looking at something like $30.
Re:GPT-6 Sol and Luna by WinstonSmith84 on Wednesday September 23, 2026 @01:46AM (#49812106)
Maybe your numbers are right, but that's not been my experience. My typical workflow is Astra coordinating with Luna Max (5.6 back then) as both implementer and reviewer and sometimes Astra review as well when I've some distrust with Luna .. A day, I've been trying to replace Luna Max by Deepseek v4.1 flash and I've been burning about $7 worth of tokens in Fireworks in a single day. More than what my 20x OpenAI sub costs me, including Astra usage. And that was when Luna 5.6 was less capable and more expensive than Luna 6.0.
Re:GPT-6 Sol and Luna by sieve on Wednesday September 23, 2026 @03:47AM (#49812895)
I have written about my experience. I have also mentioned the kind of code I write. It is not react/js/css heavy stuff that I see a lot of people write. So the code bases are typically in the 5-50KLOC range. Freestanding C, Python, or maybe some TypeScript. And fairly modular. I can thus run models on specific modules without having them read everything into context.

So the workflows I mention work for this kind of stuff.

Re:GPT-6 Sol and Luna by ascorbic on Wednesday September 23, 2026 @05:05AM (#49813401)
You can't compare a subscription to API prices. OpenCode Go is massively subsidised. Unlike the closed labs, we can say that for sure because we can see what they're paying for their tokens.
Re:GPT-6 Sol and Luna by user43928 on Tuesday September 22, 2026 @06:19PM (#49809026)
6-luna is no improvement over 5.6, merely a price cut.

And info from the help page with message limits suggests the 50% price cut does not apply to the subscription, where they applied only a 1/3 price cut instead.

I'm not thrilled with this release.

Opus 5.5, which matches GPT-6 Astra performance at a cheaper price, is much more interesting.

Re:GPT-6 Sol and Luna by oblio on Wednesday September 23, 2026 @05:36AM (#49813592)
> I dont know how they make money here

That one's easy, they don't make money.

Re:GPT-6 Sol and Luna by matznerd on Tuesday September 22, 2026 @03:59PM (#49807261)
Simon, love your work, one piece of minor feedback for the individual model pages is to make the font of the model name potentially bigger than (and above) the conversation id (which means nothing to the audience) "2026-09-22T18:28:00 conversation: 01m355zvyw8946qyraa8zpz6h9 id: 01m355zvyx47zxx5c6q6b3fg0m#".

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)...

Re:GPT-6 Sol and Luna by simonw on Tuesday September 22, 2026 @04:05PM (#49807333)
That's a good idea. It's the default output for my `llm logs` command, but that header could at least show the model ID.

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.

Re:GPT-6 Sol and Luna by matznerd on Tuesday September 22, 2026 @06:20PM (#49809032)
I think you can make it work without leaving Gists by using a Cloudflare Worker as a workaround. The Worker sits in front of the renderer page and adds the og tags to the HTML before it's sent out. You'd also need to turn the SVGs in the Gist into a PNG for the og:image, and decide if you want a grid or just one image, any text formatting, and how long to cache...

I got it working in a quick local test (grid of all the reasoning efforts, cached per Gist, loads from the raw Gist URL so it doesn't hit the GitHub API rate limit).

Code + prompt + notes here: https://gist.github.com/matznerd/ece297107bd99ac028c7962c217...

Basic concept is to:

1. Put a Worker on the /markdown-svg-renderer route. Normal visitors get your page exactly as it is now.

2. When a link has ?url=<gist>, the Worker reads the Gist and adds og:title, og:description and og:image to the page's HTML. Link previewers like Slack and iMessage don't run JS, so this is the only way they see them.

3. og:image points to a second Worker URL (og.png?url=<gist>). It takes the SVGs from the Gist, puts them in a grid, and converts it to a PNG, since previewers won't show SVGs.

4. Both results get cached per Gist, so each Gist is only fetched and rendered once, even with a lot of traffic.

Things to customize:

- Title and description (mine: "gpt-6-luna SVG of a pelican riding a bicycle" / "6 runs, reasoning effort none to max")

- Grid of all runs vs just one image, plus layout, labels and font

- How long to cache (I used a day, but edited Gists keep the old preview until it expires)

Re:GPT-6 Sol and Luna by Cu3PO42 on Tuesday September 22, 2026 @03:04PM (#49806449)
I find it very interesting that for both these models we such a clear progression of better images with higher thinking levels from 'hardly useful' to 'pretty nice'. I feel on many other models low and max are much closer.
Re:GPT-6 Sol and Luna by dom96 on Tuesday September 22, 2026 @03:35PM (#49806941)
It's surprising but MiMo V2.6 Pro performs better and is cheaper than GPT 6 Sol on my benchmark[1]. Open weight models are really snapping at the heels of the major western models.

1 - https://bench.killswitch-lang.org

Re:GPT-6 Sol and Luna by saretup on Tuesday September 22, 2026 @02:48PM (#49806212)
Not that this benchmark is super relevant anymore but these look worse than I expected.
Re:GPT-6 Sol and Luna by simonw on Tuesday September 22, 2026 @02:50PM (#49806237)
Yeah, it's interesting how much worse they are than the Astra pelicans. I think that reflects a tiny bit of genuine value still left in the benchmark, to be honest.
Re:GPT-6 Sol and Luna by hdz on Tuesday September 22, 2026 @03:08PM (#49806543)
Tons of value left, especially for open source models. I would say the benchmark is yet to be truly saturated (just look at the legs and seat to see what I am talking about) and I always look forward to seeing them. Thank you!
Re:GPT-6 Sol and Luna by Kotlopou on Tuesday September 22, 2026 @06:42PM (#49809298)
To me the main upshot of this benchmark is precisely that the pelicans still usually look a bit wonky. It's bizarre, since this definitely has a good solution, but it's in line with my experience that memorization of the training set just... isn't happening very much? As in, whether a model fails or not doesn't have much to do with whether that exact question was likely posed many times before.
Re:GPT-6 Sol and Luna by nomel on Tuesday September 22, 2026 @07:44PM (#49809794)
I think some additional value would be had by seeing how well it can modify the pelican.

Like, "now facing left", "sitting on the handlebars", or "with green spokes" to see if it can break out of some pretty obvious statistics in the training data!

And, there's always asking for an STL rather than an SVG!

Re:GPT-6 Sol and Luna by alansaber on Tuesday September 22, 2026 @03:25PM (#49806781)
It would be extremely funny if the explosion in SVG generation capability in particular was a result of this benchmark
Re:GPT-6 Sol and Luna by dmazin on Tuesday September 22, 2026 @02:47PM (#49806209)
> GPT-6 Luna being half the price of GPT-5.6 Luna is a really big deal.

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.

Re:GPT-6 Sol and Luna by agentcoops on Tuesday September 22, 2026 @03:49PM (#49807114)
I’ve been doing really heavy text analysis work with LLMs where false negatives/misses are important to minimize and my god did I hit cost thresholds quickly with 5.6 Luna — it was the first time I felt motivated to seriously work with local open models, even if inference was degraded for the task. Cheaper and much better inference now brings me back to the closed models for better or worse.
Re:GPT-6 Sol and Luna by FusionX on Tuesday September 22, 2026 @04:07PM (#49807367)
5.6 Luna was already discounted at half the price on OpenRouter. Looks like they made it permanent.
Re:GPT-6 Sol and Luna by onlyrealcuzzo on Tuesday September 22, 2026 @03:52PM (#49807160)
Hopefully Terra 6 slots somewhat nicely into this space.
Re:GPT-6 Sol and Luna by user43928 on Tuesday September 22, 2026 @04:26PM (#49807580)
Yes, I am mildly disappointed with these releases.

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.

Re:GPT-6 Sol and Luna by zigzag312 on Wednesday September 23, 2026 @04:33AM (#49813196)
Yeah me too. Maybe that place will occupy the Astra Minor model that appeared in Microsoft's Azure model config. As Sol and Sonnet are now similarly priced (unless Sonnet 5.5 will reduce its price).
Re:GPT-6 Sol and Luna by user43928 on Wednesday September 23, 2026 @04:50AM (#49813309)
Good point!

Maybe they are keeping the cheaper Astra alternative back for their Dev Day next week Tuesday.

Re:GPT-6 Sol and Luna by mkotlikov on Tuesday September 22, 2026 @02:57PM (#49806341)
How come the pelicans get older with more reasoning? Is GPT 6 taunting us with our mortality?
Re:GPT-6 Sol and Luna by zahlman on Tuesday September 22, 2026 @10:42PM (#49811006)
Probably it's easier to convey youth than age with a lower level of detail.
Re:GPT-6 Sol and Luna by psma_egeliaa on Tuesday September 22, 2026 @03:28PM (#49806839)
What's with the radial spokes? When are we gonna start seeing proper cross lacing?
Re:GPT-6 Sol and Luna by switchbak on Tuesday September 22, 2026 @05:13PM (#49808194)
And how about that head tube angle?
Re:GPT-6 Sol and Luna by arcanemachiner on Tuesday September 22, 2026 @04:09PM (#49807387)
> half the price of GPT-5.6 Luna

Half the price when it launched, or after the price dropped by 75%?

Re:GPT-6 Sol and Luna by user43928 on Tuesday September 22, 2026 @04:28PM (#49807607)
After the price drop. GPT-6 Luna does not perform better than 5.6, so they can't raise the price.
Re:GPT-6 Sol and Luna by adverbly on Tuesday September 22, 2026 @03:24PM (#49806763)
Many of them still get the layers wrong.

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.

Re:GPT-6 Sol and Luna by rayiner on Tuesday September 22, 2026 @03:48PM (#49807100)
It's funny that even Astra doesn't know you ride a bike by straddling it between your legs. (EDIT: Oh, I guess Max gets the occlusion. But it doesn't realize it has to pick direction the knee bends in.)
Re:GPT-6 Sol and Luna by pantsforbirds on Tuesday September 22, 2026 @02:44PM (#49806166)
The sol max looks like it's absolutely ripped for some reason
Re:GPT-6 Sol and Luna by redanddead on Tuesday September 22, 2026 @02:46PM (#49806189)
He’s been biking a lot
Re:GPT-6 Sol and Luna by 8bitsout on Tuesday September 22, 2026 @03:55PM (#49807197)
he's been cycling a lot
Re:GPT-6 Sol and Luna by saltysugar on Tuesday September 22, 2026 @02:55PM (#49806311)
Isn't everyone pelican-maxxing these days?
Re:GPT-6 Sol and Luna by manojlds on Tuesday September 22, 2026 @03:05PM (#49806469)
He also blogged why he thinks it's still useful
Re:GPT-6 Sol and Luna by aidos on Tuesday September 22, 2026 @03:46PM (#49807080)
That GPT-6 Sol max pelican looks… so old and depressed.
Re:GPT-6 Sol and Luna by ChickeNES on Tuesday September 22, 2026 @03:35PM (#49806935)
> GPT-6 Luna being half the price of GPT-5.6 Luna is a really big deal.

good god

Re:GPT-6 Sol and Luna by norman784 on Tuesday September 22, 2026 @03:01PM (#49806399)
Is GPT-6 50% cheaper?

> 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

Re:GPT-6 Sol and Luna by simonw on Tuesday September 22, 2026 @03:04PM (#49806464)
The table on https://developers.openai.com/api/docs/pricing is more readable:

  +--------------+-------+--------------+--------------+--------+
  | Model        | Input | Cached input | Cache writes | Output |
  +--------------+-------+--------------+--------------+--------+
  | gpt-6-luna   | $0.10 | $0.01        | $0.125       | $0.50  |
  | gpt-5.6-luna | $0.20 | $0.02        | $0.25        | $1.20  |
  +--------------+-------+--------------+--------------+--------+
Re:GPT-6 Sol and Luna by norman784 on Tuesday September 22, 2026 @03:16PM (#49806663)
Yeah, how they put, is confusing to me, they should have put that table instead of what they have right now in the article.
Re:GPT-6 Sol and Luna by tedsanders on Tuesday September 22, 2026 @03:02PM (#49806414)
Yes, GPT-6 Luna is 50%-58% cheaper than GPT-5.6 Luna. (I think the blog text and graphs make it pretty clear.)
Re:GPT-6 Sol and Luna by norman784 on Tuesday September 22, 2026 @03:18PM (#49806692)
Yeah, but it confuses me, I read left to right, so if they put GPT-6 and $0.20 first, I would assume that's the new pricing, they should make it clear, not confusing.
Re:GPT-6 Sol and Luna by dbbk on Tuesday September 22, 2026 @03:07PM (#49806514)
If you're happy with letting Meta train on you, Muse Spark 1.3 Contributor pricing is a much better deal than Luna
Re:GPT-6 Sol and Luna by addaon on Tuesday September 22, 2026 @03:17PM (#49806674)
Is Luna (on "low" thinking) the first left handed model?
Re:GPT-6 Sol and Luna by order-matters on Tuesday September 22, 2026 @03:28PM (#49806836)
out of curiosity, do you retry the same model multiple times to see the range of output it comes up with? or is it purely a 1-shot test
Re:GPT-6 Sol and Luna by hamrocksissors on Tuesday September 22, 2026 @03:43PM (#49807037)
Out of all of the benchmarks out there, pelican bicycle bench is the only one I care about. Thank you Simon.
Re:GPT-6 Sol and Luna by varispeed on Tuesday September 22, 2026 @02:54PM (#49806293)
When the Astra one was last time run? It's probably better to run these 2-4 weeks after release when models get nerfed to get idea of performance closer to what it is.
Re:GPT-6 Sol and Luna by anthonyrstevens on Tuesday September 22, 2026 @06:27PM (#49809132)
How do you know they are nerfed, and how do you know the timeframes?
Re:GPT-6 Sol and Luna by ijidak on Tuesday September 22, 2026 @03:36PM (#49806943)
What I like about the grid of SVGs is from I can see that Astra high seems to yield similar quality and price to Sol 6 max.

And Astra medium seems to yield similar or better quality for the same price as Sol 6 xhigh.

Re:GPT-6 Sol and Luna by nanook on Tuesday September 22, 2026 @03:02PM (#49806412)
Do you have a page showing all the pelicans you've ever created? Could be fun to browse - kinda like https://progress.openai.com/ but visual. (It's a shame they don't keep it updated)

I'm so tired of looking at benchmarks. I always look fwd to the pelicans.

Re:GPT-6 Sol and Luna by simonw on Tuesday September 22, 2026 @03:05PM (#49806472)
https://simonwillison.net/tags/pelican-riding-a-bicycle/ but I need to build something better.
Re:GPT-6 Sol and Luna by jdw64 on Tuesday September 22, 2026 @02:47PM (#49806204)
Looking at this, AI still has a long way to go. In Sol Max, the pelican's legs are missing on one side—how can one side have two pedals and two legs...
Re:GPT-6 Sol and Luna by loeg on Tuesday September 22, 2026 @03:11PM (#49806583)
And the bicycles have weird dimensions -- extremely slack head tube angle, handlebars in the wrong orientation, etc.
Re:GPT-6 Sol and Luna by flyinglizard on Tuesday September 22, 2026 @04:01PM (#49807283)
That's just foreshadowing the next generation of 32" all-mountain frames.
Re:GPT-6 Sol and Luna by inshard on Tuesday September 22, 2026 @04:32PM (#49807662)
My new sub-benchmark is which combinations achieve the hook at the end of the upper beak. Right now just 4: Astra Max, XHigh and Medium; GPT 6 Sol Max
Re:GPT-6 Sol and Luna by batperson on Tuesday September 22, 2026 @04:50PM (#49807889)
I've been sharing that pelican grid in my circles a whole bunch, it's great! I think only one data point is missing, generation speed. Would be interesting to see how the reasoning level/token counts relate to speed.
Re:GPT-6 Sol and Luna by viraptor on Tuesday September 22, 2026 @05:32PM (#49808433)
> Error: Gist API returned 403

Is what I'm getting on the top two links.

Re:GPT-6 Sol and Luna by shepherdjerred on Tuesday September 22, 2026 @05:47PM (#49808624)
Wow I cannot believe Luna is getting even cheaper. IMO this is the model that is going to change the world.

Everyone said tokens were too expensive but these are getting close to free while still having fantastic performance.

Re:GPT-6 Sol and Luna by redsaber on Tuesday September 22, 2026 @05:49PM (#49808652)
looks like they're positioning luna to tackle the low-cost cn models
Re:GPT-6 Sol and Luna by idk1 on Tuesday September 22, 2026 @05:59PM (#49808765)
What I overwhelmingly love about that Pelican grid is the two best ones, they've put a neck scarf on to show speed and wind.
Re:GPT-6 Sol and Luna by nicolamanzini on Tuesday September 22, 2026 @06:14PM (#49808971)
Here are some somehow standardized pelican tests but for 3d scenes in threejs at threejseval.com

Luna 6 High: https://threejseval.com/models/gpt-6-luna-high

Sol 6 High: https://threejseval.com/models/gpt-6-sol-high

You can compare any other model on the same prompt. Gallery unlocks after 4 votes: https://threejseval.com

Re:GPT-6 Sol and Luna by sfblah on Tuesday September 22, 2026 @06:55PM (#49809429)
Yep. We just switched several classification jobs we run over to gpt-6 luna. Love the cost savings.
Re:GPT-6 Sol and Luna by lofaszvanitt on Tuesday September 22, 2026 @07:30PM (#49809702)
Pelicans gonna devour capibaras if they see these depictions.
Re:GPT-6 Sol and Luna by NichoPaolucci on Tuesday September 22, 2026 @08:08PM (#49809956)
Simon - I believe you've been doing this with a "one-shot" approach. Have you ever considered seeing what the results are with a few more prompts? Maybe 1,2,3 adjustments?

Something like the astra MAX is pretty darn good - but something is up with the right wing and the right foot (flipper?)

I bet each of these could be modified to be significantly better with 1 or 2 "rounds" of adjustments. (Others not so much).

Obviously, not as deterministic as your single prompt approach, but something I just thought of while thinking about the price (Because wow! For some of these I'd expect a usable SVG after that much).

Re:GPT-6 Sol and Luna by simonw on Tuesday September 22, 2026 @09:26PM (#49810538)
Yeah, I have a couple of variants that I want to get working:

1. Each model gets three chances, and then gets to pick the best according to its vision input

2. Models run in a loop where they can produce SVG, see it rendered, and then edit it further

I tried that loop last year and had disappointing results, but the models are a lot more effective this year.

Re:GPT-6 Sol and Luna by aussieguy1234 on Tuesday September 22, 2026 @08:44PM (#49810219)
Pelicanbench
Re:GPT-6 Sol and Luna by alexforster on Tuesday September 22, 2026 @11:55PM (#49811465)
Your benchmark started being gamed by the frontier models a year ago though. The original idea (find a quirky way to test models with something they don't optimize for) is great, but it needs a refresh.
Re:GPT-6 Sol and Luna by gtirloni on Wednesday September 23, 2026 @01:07AM (#49811881)
What's the relevance of the pelican benchmark when models probably saw it during training? Didn't OpenAI stop testing against SWE-Something because it was tainted?
Re:GPT-6 Sol and Luna by genidoi on Wednesday September 23, 2026 @01:16AM (#49811934)
It's not a benchmark, it is a meme benchmark.
Re:GPT-6 Sol and Luna by a3w on Wednesday September 23, 2026 @02:47AM (#49812524)
Memes are arguably the web scale of benchmarks.
Re:GPT-6 Sol and Luna by ljm on Wednesday September 23, 2026 @07:12AM (#49814300)
AI reproducing Xtranormal video clips like NodeJS Is Web Scale should be the new benchmark.

If the dialogue is slop and not like the old memes then it fails.

Re:GPT-6 Sol and Luna by simonw on Wednesday September 23, 2026 @06:47AM (#49814098)
If they train for the benchmark, how come many of the pelicans produced by their different models at different reasoning levels still suck?

That aside, the relevance these days is in comparing models and effort levels within the same model families - hence the comparison grids.

Re:GPT-6 Sol and Luna by ksec on Wednesday September 23, 2026 @02:48AM (#49812535)
I am starting to wonder if this test is now being heavily benchmarked internally we should be using some new test?
Re:GPT-6 Sol and Luna by FranklinMaillot on Wednesday September 23, 2026 @05:12AM (#49813431)
What surprises me every time with the pelican benchmark, is that drawing style is very consistent within each model across, what I believe, are independent sessions. Same tones, similar background... Just more refined with increasing effort. I would expect much more variability.
Re:GPT-6 Sol and Luna by myrmidon on Wednesday September 23, 2026 @07:27AM (#49814422)
Damn, Astra-Max looks really good at first glance, it even has the legs on the correct side of the bike (z-order for chain is still wrong though).

I find it really interesting how consistent the layout is for these (facing right, with the sun in the top right).

Just a little more progress on physically correct z-ordering and these won't be easily identifiable as slop anymore :O

Your observation with the grid comparison is quite interesting. I wonder if that could be generalized into capturing some kind of aggregate mood/attitude for different LLMs when picking (multiple?) suitable things to compare...

Re:GPT-6 Sol and Luna by teiferer on Wednesday September 23, 2026 @09:34AM (#49815916)
> The grid is actually really interesting, because it shows that the 5.6 family default to brighter colors than the 6 family.

Could you elaborate on what it is about that observation that is "really interesting"? It is a fun detail, but does it actually mean anything for usefulness or progress or anything really beyond "gpt-6 makes darker colors"?

Not trying to dismiss your work, to the contrary. I'm wondering if I'm mising a deeper insight here.

Re:GPT-6 Sol and Luna by simonw on Wednesday September 23, 2026 @10:33AM (#49816800)
It shows that the three 5.6 models are closely enough related that they exhibit similar "taste" in their color choices, and the same is true for the 6 models.