[my002]

The era of subsidised inference is truly ending. The new model multipliers (https://docs.github.com/en/copilot/reference/copilot-billing...) seem like a huge leap, though. From 1x to 6x for new-ish GPT and Sonnet models. 27x for Opus...

Seems like folks would be better off with OpenRouter instead.

[giwook]

Lots of us have noticed that usage limits for Claude have been nerfed in recent weeks/months.

If anything, these new multipliers are more transparent than anything OpenAI or Anthropic have communicated regarding actual costs and give us a more realistic understanding of what it's costing these providers.

The fact that we were able to get such a substantial amount of usage for $20/$100/$200 a month was never meant to last and to think otherwise was perhaps a bit naive.

This feels like a strategy from the ZIRP era of tech growth where companies burned investor capital and gave away their products and services for free (or subsidized them heavily) in order to prioritize user acquisition initially. Then once they'd gained enough traction and stickiness they'd then implement a monetization strategy to capitalize on said user base.

[dualvariable]

However, inference costs for entirely good enough models are likely to keep declining in the future. We're probably hitting diminishing returns on model size and training. The new generations aren't quantum leaps anymore, and newer generations of open source models like DeepSeek are likely to start getting good enough.

There's going to be a limit to how much they can raise prices, because someone can always build out a datacenter and fill it up with open source DeepSeek inference and undercut your prices by 10x while still making a very good ROI--and that's a business model right there. Right now I'm sure there's a lot of people who will protest that they couldn't do their jobs with lesser models, but as time goes on that will get less and less. Already right now the consumers who are using AI for writing presentations, cooking recipe generation and ELI5 answers for common things, aren't going to be missing much from a lesser model. That'll actually only start to get cheaper over time.

Also for business needs, as AI inference costs escalate there comes a point where businesses rediscover human intelligence again, and start hiring/training people to do more work to use lesser models--if that is more productive in the end than shelling out large amounts of cash for inference on the latest models. [Although given how much companies waste on AWS, there's a lot of tolerance for overspending in corporations...]

[geodel]

> because someone can always build out a datacenter and fill it up with open source DeepSeek inference and undercut your prices by 10x while still making a very good ROI-

Not sure how it all works out. Currently trillion dollar companies can't make a native app for platforms. Everything is just JS/Electron because economics does not work for them.

And here companies can make GW data center running very expensive GPUs for 1/10th of current prices. Sound little fanciful to me.

[bootsmann]

The price you pay for anthropic must include the price of training new and better models which is incredibly costly. If you use the models someone else already spend money to develop you don’t need to pay this price.

[croes]

I guess the new models will still be quantum leaps, but literally: "The smallest possible change in a system"

[hansmayer]

They've been like that for a while actually, I think at least since the big hype around ChatGPT 4.5 (or was it 5?) and that underwhelming, lukewarm, oversanitised presentation by Altman and his team.

[ctoth]

Yups... Mythos is the smallest possible leap. Not a standard model generation advance, not even a version point advance. Just the smallest possible quanta of a change. We are absolutely hitting a plateau any day now. Any day. Any time. Any second now. Yup. Right now! Surely!

[hansmayer]

I mean let's be realistic - all that we know about the "mythical" Mythos is the carefully curated and release stuff by the Anthropic's PR team. Is it really a huge leap they are making it to be? I doubt it. In fact I bet if it was indeed that powerful and dangerous, as they imply, they'd find a way to release it immediately, devastate OpenAI and DeepSeek and secure a leading position in the market. Why is it not happening? I suspect because Dario is again at it, peddling his bullshit.

[cubefox]

Yeah. AI progress is insanely fast if you compare it to anything else. Where else is a one year old technology already hopelessly outdated? 10 years ago is basically stone age.

[madamelic]

I am continually tripped out by the fact when I was 16, I didn't have a 'smartphone' beyond a Windows Mobile 6 phone that had no internet on it.

Now, I have this high-resolution shiny object that can near instantaneously get any information I want along with _streaming HD video to it_ *anywhere*.

15 years even feels like a stone age. I can't fathom what it has to feel like people in their 60s and 70s.

[giwook]

I think so too.

And at some point even frontier model costs will hopefully come down (if there is still a meaningful difference between closed and open source models at that point) as all of the compute that's being built out right now comes online.

[Fire-Dragon-DoL]

I hope it's true, but right now hardware prices are insane

[hirako2000]

It does feel like the music is about to stop.

It has been years now, of cash injections, investors can't keep feeding the beast forever.

[Gigachad]

This is the best AI programming will be. From here on the enshitification starts and the prices go up.

[ochronus]

As predicted by many. The math is, as usual, mathing.

[ctoth]

It has been years now of reading this same comment... Surely people can't keep typing it forever.

[stuartq]

But the prices haven't been going up by multiples of 6 for the past few years. Things are actually changing now. I don't think it's over, but in the short term, it's going to be considerably more expensive.

[hirako2000]

They will smooth up the spike. Or be subtle and transform the existing quota so that they run out more quickly. Calling it caching, compression, optimisation, of course for the sacred benefit of the users.

That would be, even is, the smart thing to do.

[Applejinx]

And it didn't really get flawless, did it? All the same objections stand, but the cost is inevitably blowing up for the same kinda jank product.

[soraisdead]

The difference is we're now in a world where Disney has pulled out of OpenAI without comming, and Sora was dropped off a ditch.

In other words.

The bubble has burst. You're just in denial.

[hirako2000]

My read is that the bubble as burst internally (angels, seeds, VCs, and even corporate got a grasp of the inflated promise). It will take while for the actual bubble to implode.

[tclancy]

I’m not willing too, but I can set up a cron job to Claude -p the task.

[stefan_]

Dunno, if in this day and age you are making inference more expensive, more scarce, you are honestly moving in the wrong direction and DeepSeek and others will gladly take your lunch.

[Gigachad]

The hardware to run deepseek is still incredibly expensive.

[cheema33]

> The hardware to run deepseek is still incredibly expensive.

Deepseek API pricing is very low compared to Anthropic/OpenAI API pricing.

For many, the 300% difference in pricing may be difficult to justify, if the quality difference is very small. And there will be many tasks where the most expensive/the best model, is not needed. Currently many people end up using Opus 4.7/GPT 5.5 for many tasks without thinking about it.

[Gigachad]

Is deepseek still on subsidized pricing though.

[2ndorderthought]

Near zero probability of that. The model is more efficient and the company who trained it did not blunder trillions of dollars to do so. China has better electricity infrastructure than the US too, so the likelihood they can scale out before the US ever could is high. Long term deepseek, Alibaba, etc hold the most cards for sustainable AI even despite the attempted Nvidia embargo

I am not shilling China, this is just what is happening right now.

[nl]

Judging by the multiple providers selling it for around the same price (including non-VC funded competitors): no, it isn't subsidized.

[2ndorderthought]

Have you seen the news about qwen3.6? People are running it on sub 1000 euro hardware. Apparently it's about as good as Claude sonnet.

[sergiotapia]

That is folly because there is very minimal cost to switching providers, let alone models.

[bluescrn]

Did anyone really expect AI to be cheap?

If/when it gets to the point where it can replace a skilled worker, the service can be sold for close to the same price as that skilled labour. But the AI can run 24/7, reliably, and scale up/down at a moments notice.

There's not going to be much competition to drive prices down, the barriers to entry are already huge. There'll likely to be one clear winner, becoming a near-monopoly, or maybe we'll get a duopoly at best.

[hansmayer]

> Did anyone really expect AI to be cheap?

Yes, a lot of people (not me). Why? Well because that was the whole value proposition of these companies, relentlessly pushed by their PR and most of the media- rememmber it was something something Pocket PhDs, massive unemployment etc?

[rwyinuse]

"There's not going to be much competition to drive prices down, the barriers to entry are already huge. There'll likely to be one clear winner, becoming a near-monopoly, or maybe we'll get a duopoly at best."

Based on what exactly? So far every time OpenAI, Anthropic or whatever has released a new top performing model, competitors have caught up quickly. Open source models have greatly improved as well.

I expect AI to be just like cloud computing in general - AWS, Azure, GCP being the main providers, with dozens of smaller competitors offering similar services as well.

[2ndorderthought]

Right now China is flexing the future in my opinion. Smaller, widely available, frontier models for pennies on the dollar.

I think the future of ai will be breakthroughs that let it run on commodity hardware, and the average person will not be paying for it from the cloud unless they want to be surveilled or are stuck on older hardware.

Right now I am running about what was a frontier model 1-2 years ago on a junk machine. Some people are running what was a frontier model 4 months ago on PCs and laptops that cost 5,000. In a year I think the landscape will be even better.

[flir]

I do. "Commoditize your complement". Want to sell lots of silicon? Give away good local models to run on that silicon.

Even if SOTA models in the cloud are a few percentage points better, most work can be routed to local models most of the time. That leaves the cloud providers fighting over the most computationally intensive tasks. In the long term, I think models are going to be local-first.

(Unless providers can figure out a network effect that local models can't replicate).

[bluescrn]

> I think models are going to be local-first.

Why on earth would that happen when everything else is moving into the cloud to tie it to ever-escalating subscription fees and prevent piracy?

Even with gaming, where running high-end 3D games in the cloud seems like madness and inevitably degrades the quality of the experience, they won't stop trying.

[vanviegen]

> In the long term, I think models are going to be local-first.

Why? There's an inherent efficiency advantage to scale, while the only real advantage for local models (privacy/secrecy) hasn't proven convincing for broader IT either.

[solid_fuel]

Local first models aren't just more private than the API vendors, they also have the advantages of fixed cost, lower latency, and better stability - local models don't get nerfed/"updated" in the background like chatgpt does.

Maybe in a world where these AI companies behaved with some semblance of ethics and user-friendliness they would be on even ground, but for anyone paying attention local models are obviously the future.

[still_grokking]

> the only real advantage for local models (privacy/secrecy) hasn't proven convincing for broader IT either

Because of nonexistent regulation. Just wait for it…

The legal situation in for example the EU is crystal clear, only that it will take some time to go though all court instances.

[2ndorderthought]

It's foolish not to care about privacy especially as a company. You know how it prevents you from emailing yourself your tax documents? Meanwhile thousands of employees are sending literal design docs, software, product goals, etc to several ai third partys. Not only is that insane, the companies they are sending it too intend too and openly admit to scanning the data, make software products themselves, and intend to create models that can produce their products automatically.

The reason local models hasn't caught on is several fold. It's marketing to say your company follows the latest trend, and there's an inherent pressure to keep AI companies afloat so the economy doesn't entirely collapse. The other is, it wasn't until the last month that these models have caught up to frontier models. They just did, and they are more efficient and don't require a team of 500 to deploy.

[LtWorf]

To not depend on an external company that can decide the price.

[Dylan16807]

That's a silly reason. For non-agent use cases what kind of utilization are you going to average on your own GPU, 5-10%? And that's without batching.

Even with overhead and scaling for peak use and a large profit margin, any company with an ounce of competition will be vastly cheaper than self-hosting. And for models you can run yourself, there will be plenty of competition.

[soraisdead]

> Did anyone really expect AI to be cheap?

Considering most of the cost of producing a model is the upfront cost rather than the running one, I kinda still do.

The point was never to produce 4 frontier models per company a year.

[skeeter2020]

"This change aligns Copilot pricing with actual usage and is an important step toward a sustainable, reliable Copilot business and experience for all users."

I see statements like this as strong indicators that the sales people are wrapping up their work and the accountants are taking over. The land rush is switching to an operational efficiency play.

[torben-friis]

The sooner the better. Let's take a look at the long term, enshittified, viable product before we get too dependent on the trial version.

[fsniper]

And enshitification starts.

[specproc]

Yeah, totally. The recent pricing changes have just made my Copilot subscription go from great deal to awful value over night.

I've been wanting to get off MS more generally and this is good motivation. Will be playing round with OR this week.

[cedws]

Just be aware OpenRouter charges a 5.5% fee, I didn’t know until recently. I like the product, and I think the fee is fair, but if you want the absolute best pricing then go direct.

[ffsm8]

But with open router you can always just use the latest model. If you're committed to eg Claude opus then you're better off going directly to anthropic for sure, but if not, varying other models may be fine too, depending on use case and be massively cheaper. Eg new deep seek model with same mio context window or Kimi k2.6 with 270k context window for subagents which implement

[gruez]

>but if not, varying other models may be fine too, depending on use case and be massively cheaper

Do inference providers have standardized endpoints, or at least endpoints compatible with claude code? Otherwise to pay 5.5% on all your tokens just so it's slightly easier to swap providers (ie. changing a few urls?)

[swiftcoder]

> Do inference providers have standardized endpoints, or at least endpoints compatible with claude code?

Yep, you can plug deepseek/kimi/minimax into claude code just fine. Or run everything through another harness like opencode instead.

[attentive]

Or you could use gcp Vertex or aws Bedrock and still have access to a bunch of FMs without a markup.

[AntiUSAbah]

Wow thats a lot for routing traffic.

[sailfast]

And handling API tokens, and billing, and reliability, and middleware. I am not affiliated with them but it’s not “just” routing.

Apple still charges 30%. 5.5 seems pretty reasonable. /shrug I dunno.

[Dylan16807]

> handling API tokens

Don't you still need to handle tokens with them? Also that's trivial.

> billing

Yes but you'd be paying for billing anyway.

> reliability

They increase reliability?

> middleware

Which you wouldn't need if you paid directly.

I'm not saying they shouldn't get 5.5%, but that list is mostly non-convincing.

> Apple still charges 30%.

3 of the 30 is for billing, with the rest mostly being gatekeeping with a fake justification on top.

[polski-g]

There's nothing trivial about getting a Google API key. Openrouter removes that stress from my life. And I can route requests to providers above a certain TPS threshold. And much more.

[sailfast]

My point was that it centralizes this to one place instead of 10 for engineers, not that you wouldn’t have to deal with these things at all.

A single point of access with a single key for all of these things is a worthwhile convenience.

[brianush1]

> They increase reliability?

For models that have multiple providers, they automatically route your requests to a different provider if one of them goes down.

[ac29]

Payment processing likely eats up at least 2-3% of that

[arcanemachiner]

IIRC OpenRouter charges you for the payment processing fee also.

Still worth it IMO to be able to switch from Provider A to Provider B if Provider A is having a bad day.

[webworker]

I will not be renewing/switching over, either.

I had copilot mainly so I could write issues and throw agents at it, while I went off and did other things. Has been great for contained spot work.

At this point, I'll go ahead and leave it expire, and then consolidate between Codex and JetBrains AI. Especially since Xcode supports Codex with a first-party integration.

[nacs]

Even Sonnet 4.6 is 9x multiplier (previously 1x)!

The only model I even used on Copilot was Sonnet and now its got a ridiculous multiplier.

At this point they might as well just charge per Million tokens like every other provider instead of having a subscription.

[krzyk]

They do for any new plan. Those multipliers are only for people that paid annually. After their subscription ends they'll go into token based pricing like the rest of people.

[mitjam]

I understand it like : the 10 usd is for handling the business record, maybe also the harness, I get a few coins to kick tires, but to use it for anything real it’s pay as you go by the tokens list price.

[altmanaltman]

> At this point they might as well just charge per Million tokens like every other provider instead of having a subscription.

Pretty sure that's what they will eventually do

[tjoff]

... that is exactly what they will do. Just click the link in this thread, or read the headline.

[hrpnk]

Why the multipliers then at all?

[lexone]

The multipliers are there only for current annual plan customers. After 2026 its all tokens.

[MattBDev]

I thought I was smart for buying the annual plan after I graduated and lost my student plan and then GitHub taking away my Copilot Pro I got for free for being a author of a popular OSS project. Turns out I'm being punished for making that year commitment to them. I like to think I'm only a moderate user of GHCP so this is just terrible for me. I'm honestly thinking about cancelling and switching to alternatives while also looking at investing in a local LLM setup.

[fuglede_]

So they're changing the product that people already paid an annual subscription for to the worse. That's asking for legal complaints.

[ItsClo688]

27x for Opus is genuinely shocking. at that point you're not paying for convenience anymore, you're just paying a GitHub tax. OpenRouter or direct API makes way more sense unless you're really glued to the IDE integration.

[thrdbndndn]

I keep seeing people mention OpenRouter.

Does it effectively bypass regional restrictions for you, so you can use something like the Claude API from unsupported regions such as Hong Kong, or does it still enforce the official providers' geo-restrictions?

[rvnx]

OpenRouter is great for budget control, but as they are indirect APIs, your experience with cached tokens may vary, eventually costing much more than in direct depending on the providers.

You can pay with crypto though, which seems to be convenient for people under sanctions or with limited access, or if you are in low-tax jurisdiction (e.g. HK)

[jauntywundrkind]

Caching is advertised per model+provider.

That said I think few people using openrouter are actually being selective about providers.

It took half a day to get my opencode setup, was not friendly. A lot of manually cross referencing model and providers. I was actually mainly optimizing for relatively fast providers. It all is super fragile and I'm sure half out of date; I have no idea if these picks are still fast, no promises they are still the same price (pretty terrifying honestly).

I'm mostly on coding plans so it doesn't super affect me. But man is it a bother to maintain.

[schneehertz]

Even when using OpenRouter in Hong Kong, it is still not possible to connect to region-restricted models like Gemini

[johndough]

It's interesting that the cost multiplier for Claude Sonnet 4/4.5/4.6 varies so much (1/6/9), while the API cost is exactly the same for all three models.

Also, the multiplier of 27 for Claude Opus 4.6/4. is way higher than the increase in API price would suggest.

I wonder why that is.

[vanviegen]

On GitHub copilot you pay per prompt. More powerful models can do a lot more work (consuming a lot more tokens) per prompt. Also, they tend to use more thinking tokens.

[johndough]

> More powerful models can do a lot more work (consuming a lot more tokens) per prompt.

That is not my experience. Each model since at least GPT-4 can fill up an entire context window. In fact, more powerful models can solve tasks faster, so their ratio of multiplier to API price should decrease, not increase.

For example, Claude Sonnet 4.6 has a multiplier of 9 and an API price of $15, which is 0.6 multiplier per dollar.

Claude Opus 4.7 has an API price of $25, so it should have a multiplier of 25 * 0.6 = 15 when extrapolating from Sonnet, but the multiplier is 27.

> Also, they tend to use more thinking tokens.

That might be it. Is there any data on this somewhere?

[port11]

> Is there any data on this somewhere?

Anecdata: for me, this is exactly the case with Opus. It _really_ thinks, looks into more sources, more of the codebase, etc. Sonnet is 80% thorough, but Opus can go the extra mile and burns a ton of tokens doing so.

[rvnx]

One theory of the play of SpaceX might do if everyone migrates to query-based billing:

Provide cheap and unlimited access to Grok for programmers (hence the Cursor partnership/purchase for distribution).

-> This would drag massive revenue right before the IPO announcement, like if the company is super growing

-> At a loss, but don't worry, we need these funds to build the biggest datacenter of the universe.

This announcement would create enough momentum to increase valuation, and because of the merge of his companies, would save his X/Twitter investors from a tragedy.

-> Would also be a great service to Cursor investors and so, who are stuck with their VSCode fork

[minimaxir]

It takes longer to build a datacenter with that much capacity than it does for the market to respond.

[2ndorderthought]

Buying real estate in imaginary places is lucrative at first

[AntiUSAbah]

They probably want the training data. Otherwise these 60B don't make sense at all.

But they can't buy curser before their IPO so thats that?

Perhaps they have to much compute because Musk overpromised and Twittergroq doesn't need that much compute after he nerved the porn stuff?

[hgoel]

I think they're going to have to do a lot to overcome the Musk and Grok poison. Even ChatGPT hasn't had as many lapses as Grok has had.

[gigiogigione]

I don’t get the SpaceX reference. I thought they made rockets?

[victorbjorklund]

Nobody is paying for Elons xAI so he used SpaceX to buy xAI to fund it.

[sethops1]

Under the pretense that SpaceX will be used to launch material into space to build space data centers.

[vizzier]

They now also own xAI

[0xffff2]

Which in turn owns Twitter. SpaceX is now a social media company in addition to a rocket company.

One theory I think Matt Levine posited, is that SpaceX will go public with dual-class stock that gives Elon control even with a minority ownership stake, and will subsequently buy Tesla, which doesn't have dual class stock, making SpaceX the singular "Elon Musk company", with him having operational control despite being public.

[lioeters]

That theory aligns with Elon's long-held dream of X as the "Everything Company".

[Ballas]

Then he'll rebrand SpaceX to " X" (a space followed by an X).

[mullingitover]

The point of this loss leading is to properly hoover up the money in the pockets of enterprise customers, get them locked into the idea that they need the latest and greatest cloud-based model, while simultaneously starving everyone of the memory they'd need in order to run competent models locally.

In not-too-distant future we're going to be running better models on our phones than we can buy access to today in the cloud. Skate where the puck is going: soak the customers until that day comes.

[port11]

I think your first paragraph is spot on, while the second is fairly incorrect. Hardware isn’t getting cheaper at a reasonable pace, and datacenters will keep depleting the market. State-of-the-art models are very, very far from being run on your own hardware.

[mullingitover]

> State-of-the-art models are very, very far from being run on your own hardware.

Still, the models will only get smarter and more efficient as the hardware gets cheaper. The timeframe may be debatable but the outcome really isn't.

[port11]

I could accept that the end result might be what you propose, but training models is getting tricky, running them more so, now that hardware is becoming pricier. The future might simply be a few feudal lords permitting you to run the best models on their equipment, and a few good open models that most will struggle to run.

[mkhalil]

Why would folks be better paying 5.5% fee to OpenRouter ("Open") if most people just use one or two providers? Just use the provider's API.

[djeastm]

The routing automatically routes you to other inference providers (for the same model) if/when the original provider goes down.

It's a convenience cost, for sure, but it's not valueless in a fast-moving world. Certainly if you're comfortable with one provider and it's cheaper, do that.

[sally_glance]

For me the largest value-add is the unified API. Being able to instantly start trialling a new model with zero code changes is well worth 5%. The other part is not having to deal with billing for multiple platforms.

[krzyk]

Those multiplier are only for grandfathered Pro an Pro+ plans that had annual billing, basically a way to scare people of out of those plans. Ant new ones (and bussiness+enterprise plans) will be on token based billing since June 1.

[joelthelion]

Can't wait for people to migrate to open tools (opencode/openrouter). This will unlock a lot of innovation.

(I know openrouter is not open, but it allows competition and should be easily replaceable if needed)

[minimaxir]

What's annoying is that it's obvious. In the case of GPT 5.5, if Copilot is going to charge 7.5x what GPT 5.4 costs while OpenAI themselves via the API/Codex only charges 2x of what GPT 5.4 costs, that will immediately raise an eyebrow.

[boothby]

To anybody who's been watching the tech sector with a critical eye for pretty much any period from the late 90s and onward, this is just the enshittification process. For most of OpenAI's existence it's been obvious, to me, that investors were burning insane levels of capital to build the market, and now that folks are locked in, you're seeing higher fees, ads, etc. Yet again, the user is the product; the investors want to siphon your data, attention and once you're hooked, money. And for companies like Microsoft and Apple, those hooks can dig deep.

[Incipient]

I'd call it a straight up "bait and switch".

[BearOso]

If you paid attention to the power requirements and amount of hardware being put into data centers, you should have realized that it cost them an order of magnitude more than you were being charged. To rework your analogy: they hooked you, now they're gonna see if they can reel you in.

[AntiUSAbah]

They can only reel you in if its worth it. I still can code.

And while i do not spend 200$ privat, in my startup we discussed this and our current mental model is, that instead of hiring someone new, we prefer to have more money for tokens.

This is easier for us and has a bigger benefit. The cost of a new / first employee is very high, a 200$ subscription is not. Upgrading that to lets say 400 or 800$ is still alot easier and if i can run multiply and better agents with that money, lets goooo.

[boothby]

I'm looking at education -- teachers and students, not terribly tech savvy, are being mandated to use these tools. And then comes the rug-pull. It was worth it, but now it's outside of their budget. Poorer schools / students can't stay at the cutting edge; richer schools / students can.

[jochem9]

Oh, I thought it was opium.

[Ekaros]

Let's call it for what it is dumping. Dumping things on market below cost of production. This should not have ever been allowed. RnD costs I can accept somehow. But in this case the interference should have always been billed for the real costs that it took to produce and pay off the capex.

[Gagarin1917]

“Enshitification” is just when unsustainable subsidies end?

Another reason to hate that word.

From a different perspective, you were granted an incredible gift from the companies who let you use their product on their dime. Hopefully you made the most of it when you had the opportunity.

[boothby]

No, it's much more than that. It starts with unsustainable subsidies, as Uber undermined the taxi industry with a ludicrous burn rate. And then, once everybody's hooked to the point that they can't imagine life without the product, you raise costs. And you iterate: raising costs, lowering quality, selling data, increasing addictiveness. Until everybody wants to get rid of it, hates every aspect of it, but is still hooked to the core product. I'm personally not using these tools, not using uber or Meta products. But I'm still using some Google products and it's hard to extricate them from my life now that I'm using them.

[Gagarin1917]

> No, it's much more than that.

Okay then this AI stuff isn’t an example of that even under your definition.

[itemize123]

unless the 5.4 price is a huge loss leader for them

[recitedropper]

Everyone seems to believe OpenRouter isn't subsidizing but, until they publish audited financials, I personally doubt it.

[deaux]

OpenRouter doesn't even have hardware. What are they possibly subsidizing? The platform costs?

OpenRouter is guaranteed to be about the highest margin operator in the business right now. Everyone wishes they'd be them, skimming 5% off as the middleman without any OpEx.

[ValentineC]

> OpenRouter is guaranteed to be about the highest margin operator in the business right now. Everyone wishes they'd be them, skimming 5% off as the middleman without any OpEx.

The 5% fee probably has to factor in Stripe's fees, which would be around 3% to 4% depending on whether it's an international card.

[recitedropper]

Streaming, caching, and tool calling can get pretty expensive with scale, even when you don't touch inference. Maybe they're doing something clever and are quite profitable.. or maybe they've already taken $40mm from VCs and are currently trying to raise $120mm at a 1.3B evaluation.

They also show headline prices for the cheapest provider of whatever model, but then need to hit different backends some of which may be more expensive. For now they absorb those costs, but the VCs always come knocking.

Just my opinion though. Totally agreed that they have one of the best positions amongst all AI providers from a financial standpoint.

[vanviegen]

> They also show headline prices for the cheapest provider of whatever model, but then need to hit different backends some of which may be more expensive. For now they absorb those costs, [..]

They do?? I was under the impression I was just playing the price for whatever provider they deemed 'best' for each completion.

[recitedropper]

That is what I had heard.

Checking now: The way they describe it in their FAQ is that if the price changes, then they will bill you the new price. But I read that as regarding if the primary model provider changes their headline token cost; not in the case of pricing differences for models that have many different backends that host them.

Regardless, I would be more concerned about the streaming costs if the service continues to blow up and they scale aggressively through VC investments. If their 5.5% skim accounted for what they needed, you'd think they could effectively grow organically..

[Mattwmaster58]

FYI, these are the multipliers for annual plan. I would hazard a guess most people are not on an annual plan

[mitjam]

I am and I see it as stopping the music at a party when you want everyone to go home without telling them to go home. There is also the offer to quit with prorated refund for the remaining time. I think I am going to take it.

[sandos]

Wow, having a corp. account I do wonder WHEN we are getting some kind of resctriction of usage, or require us to justify our usage.

That GPT4-mini change is going to be brutal! Its much better than 5-mini, which was itself much better than earlier free models.

[youwangd]

Show HN timing matters more than people think. Monday-Thursday, 9-11am Pacific, is when the front page has the most engaged readers. Weekend posts get less competition but also less engagement.

[themafia]

We can't even get slop delivery worked out. So use SlopAggregator instead.

[darqis]

IT'S NOT SUBSIDIZED

[whateveracct]

"eras" tend to not be so short lol

[2ndorderthought]

I don't know if it's just me but copilot kind of sucks. I've been running local models with like 9b parameters and they are about as good if not better. Obviously there's no integrations or whatever and I get most people are probably paying for that than anything else but eh. Big no thanks from me.

[siva7]

That's so unfair to us hard working developers. A month ago i could buy for .4$ a turn with Sonnet. Now i have to pay at least .9$ for this turn. Weeks ago i could buy for .12$ an Opus turn after they already raised prices and now they want .27$ from me for the same product! They are stealing from us!

[asdfasgasdgasdg]

They aren't stealing from us, for several reasons. First of all, it's a voluntary transaction. If you don't like the prices, use something else. Or don't use AI at all.

Second, you have no idea what their costs are. It is most likely that they are simply passing on their costs to you. If that was not the setup, users would just go to another service provider who was providing tokens at a cheaper rate. It's not like there is a dearth of competitors in this business.

[croes]

The already stole when they trained their models on the data.

Now they just increase the price to buy it back