OpenRouter is joining Stripe
openrouter.ai
Previously: Stripe will reportedly acquire OpenRouter for $7B+ https://news.ycombinator.com/item?id=49323381
openrouter.ai
I just want to point out some features of OpenRouter that make it more than just a model selection and routing endpoint and that I find incredibly useful:
0/ Default routing is to the cheapest provider, but they're usually not the most performant. I'd guess 99% of OpenRouter integrations never tweak the default routing. Here you can setup cheapest with performance minimums:
https://openrouter.ai/docs/guides/routing/provider-selection...
You can also stack model selection in priority
1/ Using broadcast you can push all your analytics to clickhouse / s3 / snowflake and a bunch of other compatible destinations. Setup a clickhouse server ($5 VPS[0]) and send all your traces to it:
https://openrouter.ai/docs/guides/features/broadcast
customise your own observability in your dashboards from there. Superwin
2/ Model router is also a natural home for llm security - OpenRouter has the beginnings of prompt injection detection:
https://openrouter.ai/docs/guides/features/guardrails/prompt...
there is also PII detection. This will show up in observability as rejections/blocks etc.
There are so many model routing solutions (same with observability, security etc.) but they're all 80% solutions - OpenRouter really rounds out with well implemented features that you need when deploying models at any scale and I gladly pay the toll.
[0] not sure if these exist any more but clickhouse is resource efficient
But don't worry, many providers there charge for cache the same price like non cached ;)
I want to know which vendor/model does best at my extracting-facts-from-text task? Which does best at my OCR-a-text-document task? Which can deal with a safe-for-work beach photo without a censorship system false alarm?
OpenRouter lets me run my tests against openai and anthropic and google and x and bytedance and qwen and llama, with a single sign-up and a single payment.
All those features show it's just a model selector and router.
Maybe I'm biased from the perspective of a "harness provider", but I think Model routers often have too little context to act as well-informed prompt injection prevention. e.g. it lacks context of where which part of the message(s) originates from and sanitization/safeguards were already performed on the application layer.
Something like OpenRouter's "flag" mode is fine, but usage of auto-redact or auto-block should really only be used if there is significant risk exposure through your harness or otherwise they are a constant source of bugs.
And if they are, their compliance team is about to strike them down. The VCs forcing this acquisition do know this.
- Why would you add a penalty of 50 ms at a minimum? And that is not the p95... Just run LiteLLM in house and you dont even really need that.
- Their capacity pools are shared across the whole user base, a massive batch processing by another of their customers and think what that means for your response time...
- So instead of negotiating corporate rates with OpenAI or Anthropic, you would be using an intermediary and topping up the corporate credit card... for a 5% markdown ? Really?
- They can see all your critical corporate data on the in and out
- They present some pink SOC 2 promises but then wash their hands and defer to you and the providers. Its just the Bolt and Uber model the drivers are not our employees....
- They are a man in the middle proxy that is a massive security liability for your corporation
- They have no support for private cloud points
- No geofencing guarantees
- No intellectual property legal indemnification unlike what AWS or Microsoft or Google offers
- Its a provider roulette inconsistent with hosts providing different quantization levels causing random shifts in response quality
- Support via a Discord server...
The only reason they were not shutdown yet by Anthropic or OpenAI is because they have the same VCs, as those two. That would mean said VCs investment would go to zero. Oh and those are the same VCs that own Stripe...
Just setup a private proxy tier using something like LiteLLM, even if you really dont need it. Just code your enterprise apps to have have fallback loops on the core hyperscaler providers like AWS Bedrock or Azure Foundry...
All of the customer service bots I've dealt with recently were reasoning models that thought for multiple seconds. I would not have noticed 50ms extra.
I think the bet is that you offer enough value-adds that companies who find traction don’t want to switch to something enterprise grade, and you grow into being enterprise grade with them.
Very similar to Stripe’s trajectory, incidentally.
Turns out even a proxy can be worth $8bn with the right business model behind it.
Users get an array of providers competing behind a single API, meaning they have to compete on price and quality not vendor lock-in. This encourages users to join OpenRouter over specific model vendors.
Providers get easy access to revenue (and data) and new customers with little to no ad spending, encouraging them onto the platform too.
And that's all you need. Win win.
Well done and congratulations.
An underpinning of their model is that API calls / inference are similar across providers allowing for commodity tokenization cost comparisons, but the providers are beginning to shift to non commodity features that don't easily shift.
For instance, calling Gemini with "search grounding" isn't something that OpenRouter can do (they sub their own web search in), last I checked they weren't doing real time voice models, etc.
It can: https://openrouter.ai/docs/guides/features/server-tools/web-...
It's apparently even the default these days (which makes sense, as it's usually better in my experience).
Another example might be all the managed agent features, where you get a VM + model, right now it's mostly Google and Anthropic that have this offering.
The service tier is another, where Google, Vertex, OAI, Anth offer it but only OAI and Google offer flex service tier and not all models from them get it.
OpenRouter for me is an abstraction over that complexity, they will have their work cut out for them.
They are fighting to be the one and only
The smaller, less known could get a win, but the bigger lose.
They make less than 140 million in revenue, what today is like 5 signing bonus for OpenAI :-) They are not profitable and see through less than 2 billion in revenue.
These are easy to check so...
I also congratulate the founders for pulling this off.
People said a similar thing for Cursor simply being a VSCode wrapper for when it crossed unicorn status.
I really think they only know how to sell Azure.
No reason Microsoft couldn’t have shipped a better version of Cursor considering they wrote the IDE cursor is based on.
Stripe and every other big company could build this without any issues.
It either is just a really stupid business decision or it is about the name. The only model proxy i know is OpenRouter despite plenty of other model proxies existing.
I wait for the moment where engineers realise there's more to a product than just the building aspect.
Nonetheless, just because someone payed that much money for it, doesn't has to mean antyhing.
likely piggybacking on the Open in OpenAI which is also not open source
Stripe can use OpenRouter to build the financial and accounting infrastructure for every product that sells metered AI work.
I think the analogy is ADP. Payroll for all the work that's going to be done by AI agents.
Perhaps stripe thinks metering is a lovely big market? A nice complement to their existing service. Plus a little AI buzz likely helps their valuation. OpenRouter could be a step rather than a goal.
The user could log into his personal OpenRouter account, authorize your application, and optionally set a budget, all in something like a Stripe payment screen.
Like standard oauth?
As AI agents/harness/human are spenders of tokens, enabling them to derisk from being locked to a specific model provider & allowing to (re)route to any model at anytime for better leverage in a single API, similar to how they are doing for payments.
Example: access to real-time stock trading data, access to weather information, letting it make stock trades, etc.
At first I was confused why Stripe bought OpenRouter, but I think this makes sense.
Which is saying a lot.
And it's a lot more likely that LLMs can't be changed to become unreliable, it's just how they work. So we would need more basic research, that doesn't grow on trees and for which the timelines are basically open ended. Maybe tomorrow, maybe right after cold fusion hits mass adoption.
And there is no necessity. We have accountants.
Also, how many accountants do you know that are truly happy with their work? I'd like to think many of them would love to do something other than crunch numbers all day.
We're far removed from "we've sold all the computers the world can buy" (back from when computers were the size of a building).
This is a disingenuous argument. The web and email were basically instant hits and people realized it. Similar story for home computing once the computing power caught up.
Lots of techies are tech-optimists ("tech always improves quickly").
Lots of people also have dollar signs in their eyes (or related, such as increased visibility and scope).
Hard to tell which is which.
I ask because if u were to start speaking about the specifics, might we be able to discuss some specifics from that world..
Many of the transactions are a series of small actions, not big things.
Knowing what to trigger, in what order feels a little plausible.
Since most accounting systems have an audit trail, not much that can’t be worked forwards or corrected.
As user agents become more common it's only natural that they will be used for taking the heavy lifting out of e-commerce purchases. There will be a big need for digital payments to verify and reconcile these purchases.
With LLMs that are plugged into digital payments we will essentially have buying agents in our pocket that can find us exactly what we want for the cheapest price and the quickest delivery.
Truer words, never spoken. I'm not sure how exactly this will screw me over -but I do know that it will.
https://cortecs.ai/detailedServerlessView/deepseek-v4-flash-...
https://openrouter.ai/deepseek/deepseek-v4-flash-0731#provid...
It is though, if that is what you want
YouTube there’s no real alternative because hosting unlimited amounts of video for free is a money pit that’s impossible to turn into a viable business.
If some big corporation wasn’t willing to subsidise it for some ulterior motive, it simply wouldn’t exist
They had no true reason to sell beyond greed?
I'm aware that OpenRouter checks response quality onboarding, and does further checks occasionally, but I'm concerned that it's basically a cat-and-a-mouse problem between the scammers and the detectors. For example, there could be a signal that a specific pattern of requests are from OpenRouter's quality testing bots. Or, they can just route 1% of requests to an inferior model and benefit a small gain, hoping it fits into the statistically allowed margin.
They also support fallback by default so you don’t have to write wrappers and logic to choose models, it just works with their SDk using config.
And if it's not worth it, then are you spending enough where it even effects Openrouter's bottom line?
With enough devs playing with hobby projects I'm not doubting it's profitable, only that $7 billion seems way too high.
It also has way better uptime than the underlying platforms, even for proprietary models like Claude. When Claude APIs are having issues, OpenRouter Claude still keeps working because they can route to AWS Bedrock instead of Anthropic etc. This effect is even bigger with open-weight models because they typically have 5-10 providers.
The good thing about the developer experience is that it is dead simple to create keys that are time-bound, model restricted, and have budgets against them in an easy, programmatic way (also in bulk). Costs are guaranteed to be on par or lower than any model provider, so it's an easy sell to the finance department.
To look at it from the other side, Gemini is the absolute worse when it comes to key management. The labyrinth of actions that one has to navigate to generate budgeted keys for a team is not worth it and I still haven't figured out how to manage the keys at a granularity I need.
OpenAI and Anthropic both have what I would consider to be middling DevEx. Most things work, but only on their platform, and there are some weirdnesses, like the fact that anthropic disabled its admin API key creation, so you can't programmatically create keys against a budget.
I saw another comment thread on this recently. What I gathered was they hid it in the Cloud Platform, so for those who jumped the cloud hoops already, maybe it's alright. (However when I read what Openrouter offers, as you just laid out, I feel Google can't hold a candle to Openrouter DevEx.)
However, if there's anyone who is aware, was there a brief point where getting a Gemini API key was simple (enough)? I'm asking because, in true Google fashion, the AI product went through a lot of changes.
You would think that these things would be standard fare, but they really are not. Other providers need metadata from a separate catalog, don't report pricing, don't really report available credits / usage either.
So yes, OpenRouter does have a good DevEx.
On one walk I asked 'why are you doing another startup?'
For context, his last one, OpenSea, was valued well into the billions, so it wasn't for money.
His reply: "I just love solving all the puzzles."
It's incredibly hard to compete with someone who is playing the game for the love of the game.
Kudos on playing well, Alex.
DOn't get me wrong, great for him to make money and being able to move fast and succeed, but you could play this game to if you want.
Using something like OpenRouter (or any of the AI Gateways) is better than tying yourself to one LLM provider that can rug pull on pricing or change models in a way you don't like.
The value is in the network effect I think. OpenRouter is popular and has a good head start over anything Stripe could build internally.
No one is talking about trading tokens.
The idea being that getting your tokens from Provider A is no different than Provider B, especially if they both offer the same model. You can change one value in a request to openrouter and suddenly be hitting a different provider but offering the same tokens, because they offer the same model and the same settings.
You're describing a commodity. However, the commodity isn't the tokens, but the compute capacity, i.e., serving a model. And compute capacity isn't a currency — at least not, until you can acquire compute capacity from one party and exchange it with another party.
dgellow is absolutely right: tokens aren't a currency, nor can you trade them, nor are they fungible. The original claim that "tokens are just a type of currency" [1] makes no sense.
We are basically back to 2020 trying to get people to understand that an NFT is NOT the underlying asset it abstractly represents…
Please go on as I do not get your point.
Tokens don’t have an inherent value and have no other properties that would make they tradeable.
Another way to think about it: a company that spends $1M on tokens is burning that money hoping they will be able to recoup by generating a better product. If it would be tradeable, they could spend $1M to generate tokens, produce absolutely no product, and just resell all those tokens to get their money back. That’s obviously nonsense, that’s not at all what tokens are. And they cannot transform the tokens they got into compute, you need to consume hardware and energy to mint tokens, you cannot convert tokens into hardware
Can’t explain the valuation since everything in this space is rationally overvalued, but I don’t think OpenRouters valuation is that surprising, all things considered.
I can imagine just the opposite outcome from the same scenario: as people settle into their favorite but commoditized models, competition for marginal inference cost will take over. A company like OpenRouter that promises the cheapest tokens by the minute becomes essential on the low-cost margin.
I think that OpenRouter and equivalents get pushed out of the market only if the froth calms down (as you posit) and winning models stay proprietary, perhaps with their own unique API surfaces.
Okay, I can see that, but if their value is just cost optimization, their ability to mark up the tokens becomes increasingly difficult as well. Or, people will build a router themselves to avoid paying the markup, possibly with reduced features, but someone will open source it. Heck, Claude or GPT can probably one-shot it these days. Either way, I think the whole OpenRouter model is going to struggle unless the market stays frothy.
Also, you don't need to mark up tokens much if you're a commodity volume business. Think of Costco and their margins & membership fees. Not everything has to be high margin, not everything has to be a SaaS subscription.
The other thing is convenience and centralized security from using one gateway to access everything. It's a lot better than having to deal with N accounts with separate limits and monitoring. And giving your payment details to one company instead of 20 is obviously safer.
Investors like them because the pricing is inherently usage based so there's zero risk of clients using more tokens than what they paid for. Guaranteed profit as long as they can keep a modest amount of customers.
If compute is constrained and expensive, OpenRouter is what you'll use to get around the constraints at individual providers.
[0] https://x.com/dwarkesh_sp/status/2084333160075055122?s=20
I don't see how stripe adds any value here (and I've had such terrible experience with stripe automatically breaking my stuff I am worried I can't trust openrouter now) but I can see stripe wanting to be in the middle of any two people giving each other money on the internet and this is squarely (lol) that.
Just guessing. The frenzy around popular, good already, and successful services with the corporate crap flowing from this acquisition announcement too is appalling. The "what's best for you, the user" heavy emphasis when this would be inherently evident in any honest service forecasts the opposite.
Some highlights from one of my agents asked for a no bullshit evaluation:
"By buying OpenRouter, they own the routing layer that decides which model gets called and at what price."
"Stripe wants to be the economic infrastructure for AI — meaning they want to be the toll booth and the traffic cop for the entire AI economy."
"... insider market intelligence that OpenAI, Anthropic, and Google don't fully have. Stripe would now have it."
"The $7B+ price is absurd on any financial metric, but this isn't a financial acquisition — it's a strategic positioning play to own the platform layer of the next computing paradigm."
"140x revenue multiple: At ~$50M revenue, this is not a financial valuation. It's a strategic land grab — buying the chokepoint before someone else does."
I do not feel a particularly strong smell of 'best for the user' here for some reason... More like the usual 'how do we squeeze out more for our PE folks from this' kind of scent.
(It was more like quoting a point of view of some source, named, btw. As you can clearly see. You have my thoughts there too. In case you missed those.)
It's an administrative burden to sign up very every model provider, and there are many independent inference providers now that serve only open source models.
OpenRouter provides a useful service by allowing easy prepaid model access with much higher rate limits, and they also aggregate different model providers to route queries by price, latency, etc.
Just because some Higher Ups in Stripe bought this, doesn't mean its that valuable.
It can easily be, that Stripe can just afford it and think that they are part of the big boys and thats just what companies cost today.
It could be that Stripe really really like the brand name and all the positive notion of it.
It could be that Stripe doesn't want to build this from scratch in a timefrime of 1 year or 2 because Stripe might be too corporate to be able to do startup stuff.
Stripe has quite a high motivation to leverage agents thoguh because they are preparing for Agents which will buy through stripe. They already provide the SKU backend and support the agent payment stuff.
Surely stripe if anyone have learned to harness the cash flowing through their system. Hell, they could be emitting bonds on expected token consumption bills!
It seems strategically not in their interest.
And OpenRouter can steer traffic away from those labs over to their own model.
The major labs don’t get personalized data about the user to train on, since it’s aggregated.
Which is just some of the reasons why strategically it doesn’t seem in their best interest to allow a middle person.
> OpenAI/Anthropic/etc don’t establish the direct customer relationship.
But they also don't have to pay to support and maintain the direct customer relationship and all the billing, support, liability, etc. The credit card merchant fee (a not insubstantial % at scale) is paid by Openrouter (or Stripe now) and the bill Openrouter pays to OpenAI is transferred probably via wire or something with a lot less fees. Meanwhile if you're paying OpenAI via credit card directly, OpenAI is paying Stripe or whoever.
> And OpenRouter can steer traffic away from those labs over to their own model.
The inverse is also true. If you've deposited money on a Openrouter account, you've taken a very big step across their proprietary moat. Would the proprietary companies rather lose them as a customer forever, or be able to entice them back, which can be done with a single change of a JSON parameter?
> The major labs don’t get personalized data about the user to train on, since it’s aggregated.
(Edit: Looked it up, Openrouter does send a pseudonymous but stable hashed userid upstream, stable across different API keys in the same account [0]. That makes de-identification a lot easier.)
First, a lot of the corporate use is getting ZDRed, so there's nothing to track without going back on zero data retention.
De-identification was trivial for normal web traffic before LLMs. Think of all the crap people's harnesses and claws send without de-identifying. Many coding agents can't help themselves from scanning your home directory.
Sure, the type of person to comment on HN probably has good hygene, but the typical vibe coder doesn't even know what they're leaking.
[0] https://openrouter.ai/docs/cookbook/administration/user-trac...
And per the rankings [0] 18.2% of the tokens sold through went to OpenAI [1]. 3.8% for Anthropic though. Google's up there at 23.4% primarily because GCP/Vertex/whatever they're offering is a billing nightmare and the only way to set a hard cap and sure you won't wake up the next day with a $10k bill is via OR.
[0] https://openrouter.ai/rankings#market-share
[1] Although gpt-oss 20b and 120b are indexed as 'openai' because they're the creator even though they don't serve gpt-oss via openrouter. But neither are in the top 20 of models this month, while Luna and Sol are.
> **Google's** up there at 23.4% primarily because GCP/Vertex/whatever they're offering is a billing nightmare and the only way to set a hard cap and sure you won't wake up the next day with a $10k bill is via OR.
OpenAI absolutely lets you set a hard or soft cap. I've set them and they've worked, I think one time I saw a $-0.01 balance. Their docs were probably written by ChatGPT. Later in that article it says:
> A spend limit can monitor spend without enforcement, or you can enforce it as a hard limit so API requests fail after spend reaches the limit.
And they still allow the most certain limit, which Google does not: have $0 on your deposited balance and don't have auto-reload enabled, you get 4XXed instead of going into debt (or at least by at most a couple cents).
Although for fun, go ask Gemini how you can set a cap on GCP spend. It'll try to write you a pub-sub client.
* getting small cut out of queries is better than getting zero cut, because people just make do with open models instead
https://news.ycombinator.com/item?id=34423387
are there more examples of this? a bit of an HN anti-portfolio. good reminder to others.
But such is normal with normies providing startup critiques, if you just look at every announcement and say "this will fail", you'd be right 99% of the time.
So nothing really has changed in this community's vibes.
I mean - just looked at this downvoted comment essentially saying the same thing - https://news.ycombinator.com/item?id=49119464. And also the intensity of the replies.
love that Dropbox has became textbook example of viral marketing
(i know much of it is a crap shoot and easy to quarterback on Monday morning.)
One great example is Open Banking. I would support an Open Router that mirrors Open Banking, unfortunately the Open here doesn’t carry the same connotation.
Kudos to the team for succeeding in executing around a clear DX issue during the chaos of providers, though long term I am not really a fan of it staying as is even if Stripe gets to be the one carrying it forward.
Open banking is slightly different, as it's more about account access (although with read+write access, you can kind of build it into a payment scheme as well, although it's usually clunky).
The brazilian central bank has a page about it[1]
[1] https://www.bcb.gov.br/estabilidadefinanceira/openfinance
[1] https://www.mckinsey.com/br/~/media/mckinsey/locations/south...
Openrouter is easy, reliable and performant (like Stripe).
Just the other day, I tried to have openrouter route to my local ollama, and it refused. Apparently I have to pay for this privilege. How wonderful.
Open open (op)en. Open open (op)en. Clo-o-o-ooooo-sed.. (sing with Wagner flight of the V)
That's Google's real power. Works for them. Between search, ads, and the "app store", they can crush most companies. OpenRouter's power is only in one area. For now.
Counterpoint: you can't just go to Visa/Mastercard or a merchant acquirer out there and set up an account on the same terms that Stripe can.
On the other hand, you can sign up to any LLM provider and get API access on terms that are the same or better (since I'm sure they don't appreciate having a middleman and would benefit from incentivizing direct usage) than OpenRouter gets.
Their only value comes from the fact that the currently have lots of traffic. And I dkn't think that their cumstomers are really bound to theur servuce. They could switch to a competitor without too much hassle.
There is a reason why not doing your own payment processing is a thing.
OpenRouter may have some interesting things in streamlining the process of switching LLM providers, but it is indeed something easy to replicate in comparison to payment processing.
Or if one of the provider suddenly decide to forbid openrouter from using their api. Why would they do that tho ?? Well it's not like execs never take dumb decisions.
In my humble opinion, it's extremely overpriced. But then, it's just the standard with AI currently. Divide every ai company valuation by 1000 and you might get it's real value.
OpenRouter is commodity stuff, I've never used it personally and picked alternatives. The sense I got from people who said they used it was that they are on average penny pinchers. That does not seem like an ideal user base
Payment platforms existed for years before Stripe came along. They solved the problems the others couldn't.
On the other hand, I operate an app that talks to (and logs prompts/meters costs) to many different LLM API providers, and I do not consider it painful at all. I have an AI agent to deal with any integration quirks, if needed. Mostly they provide OpenAI-compatible APIs anyhow. It's basically a no-brainer to go direct with the providers and save 5%, the great majority of the cost of an AI-powered app is no longer dev time implementing the integration, it's the tokens themselves.
Now is that enough revenue to earn back 7B.... I'm not too sure. I guess they're betting realy big that these sorts of model routers will gain explosive token usage in the future
Also, TrustedRouter's website is full of slop, which does not inspire much confidence.
Stripe has presumably only purchased them because they think there’s consumer-surplus to monetize here, which presumably will soon be giving me lots of reasons to use something else.
When I build an application that uses LLMs, it's tuned and tested for a specific model only. LLMs are not really drop-in replacements of each other, even different versions of the same series.
The pertinent question isn't why OpenRouter but why a router company is worth 7B.
In this case I imagine that OpenRouter's moat is going to be that businesses that already use Stripe will have a lower bar to choosing OpenRouter to provide the AI access, and clean integration with Stripe to pay for stuff like AI support bots.
The flexibility is kind of what I'm pointing to, if I understand you right. Instead of dealing with Anthropic, OpenAI, XAi etc directly and having systems to handle user assignment, budgeting etc for each, companies just deal with GitHub, which they've probably been dealing with anyway.
edit: Inflation does not account for the observed growth.
Negotiations start when sales is done.
These were highly capable negotiators.
But most likely also good salesmen.
Because after PayPal, Stripe is the biggest graveyard of frozen accounts who would be otherwise perfectly fine, but some AI decided that this card from this country at this price = send the account to null.
So while before you had comfort of using OR, right now you will have Stripe emps and their AI overseeing your queries. And don't get me wrong - Stripe will love you! But once someone pulls the lever - good luck getting at least explanation why they had to boot you out. I never heard of an account being reinstated. This will be like this, only your queries instead of CC charges, will be inspected.
I don't see how this is beneficial to anyone, esp. OR users. RIP OpenRouter.
- They support 400 different models, 80 different providers, and an unlimited number of new custom agents. Every single model, provider, and agent, has its own weirdness that has to be accounted for. Tool calls change by model. Effort changes by model. Backend APIs (messages, responses, etc) change by provider. There are thousands of specific tweaks, fixes, hacks, that need to be implemented to make this thing "just work". And you have to keep updating it all, weekly.
- On top of that, they support providers running in multiple countries, which increases the legal, logistical, financial, and networking complexity.
- On top of that, they need to figure out pricing and deals with providers, and ensure the providers allocate the necessary networking and compute capacity. They need to work with providers directly to troubleshoot and fix all kinds of issues, from networking to application.
- On top of that, they implement multiple kinds of request routers. An auto router to route your request to the best general model, a fusion model to attempt the request on multiple models and pick the best response, a pareto router to route requests to the best coding model for your request.
- On top of that, they build custom features that businesses and users want, like Data Loss Prevention (which I'm blown away they actually provide for free). Lots of very useful business functionality for managing not only what model and provider to use, but also limits on usage, filters, etc. They also implement SSO, prompt injection guardrails, logging/auditing, workspaces, etc.
- On top of that, you have to be very good at just implementing HTTP APIs. Most people aren't good at it. API design is hard, HTTP is way more complicated than it seems, network traffic shaping is a black art.
- On top of that, running applications and infrastructure, and scaling it 10x every year, is a subtle yet critical skill. No matter how good your code is, it's pointless without working hardware/running apps.
- On top of that, they provide a generous free tier, which has to be subsidized, and is probably only partially subsidized by providers.
Why pay for this rather than build one? Because there's no sense in building one. There's a reason restaurants don't build their own factories to manufacture their own pots and pans. Their business isn't selling pots and pans, it's selling people cooked steak. It's a bad idea to waste time, effort, and money trying to make things you aren't going to sell.
It's a business line more suited to finance, law, sales, and accounting people than tech people and a pretty laborious one. That's often the case when the tech looks "easy" but the sector only seems to have a few big winners.
This is some high level meaningless corpo speak.
This version is dumb/empty, and reads like brown-nosing Stripe.
Also corporate speak. Their mission is take their cut of the Internet's GDP. Nothing against them, they provide real value for that cut. But that's what the actual mission is.
You might not believe them, but that’s what the words are supposed to mean.
It's corporate bullshit not because I don't believe they can grow the pie, but because that's not their motivation.
In fact, a nice formula for coming up with your corporate mission statement would be:
1. Ignore your company's ground-truth financial incentives
2. Find a positive 2nd order effect E (or a plausible one) that you don't really care about, but maybe you genuinely think is nice
3. Claim your "passion for E" is what drove you to start your company, and keeps you working, because, of course, all great companies must have a "deeper purpose"
https://stripe.com/annual-updates/2025
> Stripe last said businesses on its platform generated $1.9 trillion in payment volume in 2025, up 34% year over year.
More transaction volume = more enterprise value potential. More GDP, total rake goes up of said economic activity they facilitate. Middleman Moat, accumulating volume (present via Paypal, future via OpenRouter).
Hermes Agent is AI slop "ai assistant" software that is being entirely developed by a bunch of AI agents. They are stuck in a constant whack-a-mole bug fix session because the agents keep breaking things.
Openrouter is being entirely propped up by garbage-tier software that is only popular because it's being endlessly hyped by idiot youtuber AI Agent Bros who are so uncreative and incapable of critical thought, they're using AI agents to tell them what content to make.
When people finally realize how bad Hermes is, OpenRouter is pretty fucked. And then when all the weebs writing furry porn get bored and move on, they'll be fucked even more.
What's really bad about this: Stripe is so tightly tied to the credit card industry that said industry's obsession with morality policing will force Stripe to start censoring openrouter.
This seems mind blowing, but the big boys seem to be behaving as if it's directionally true.
If compute is constrained and expensive, OpenRouter is what you'll use to get around the constraints at individual providers.
Note: this would include both the request, the response, and competitive responses. This would be very high quality training data.
Now at the current point they're sporting and make this data sampling option opt-out, and indeed they do. By default, they collect your data.
Model companies will pay a lot for this, especially now, more than customers will ever give to openrouter. Think about this: this is a constant stream of data of exactly what people and companies do with AI models, for all models, updating LIVE.
Second, they can provide "competitive intelligence" as well, telling models live how well they're doing and what exactly their strong and weak points are, and in what distribution of token requests this results.
There may even be financial companies that pay for this data.
Why invest in an LLM API when you are a financial service provider?
I can’t see any obvious reason to use Stripe more because of this new product.
The only hypothesis I can think of is that AI token could be a ”currency” in the future. (Which doesn’t make sense when everyone will have local LLMs)
1.Routes the request to the appropriate provider (OpenAI, Anthropic, or Google).
2. Returns the model’s response to your application.
3. Records token usage for billing automatically.
4. Rejects requests when a customer has no credit left (when enabled for your account).
Does that mean all api calls to providers go through stripe?
Reading the justification for how much of a perfect partner stripe is for openrouter sounded made up, this makes it even better.
Models are already paying per search for gated access to higher accuracy info (this is abstracted by anthropic, chatgpt etc currently where they have cut deals w Google etc to allow their traffic through)
Models don’t currently interact with the economy directly in a real way (by eg themselves deciding to pay for a high quality information source), but they will within a few months as folks get more claw-like agents
Those transactions can happen with bitcoin…. or they can happen on stripe
At least that’s my read
Stripe wants to take a cut of as many transactions as possible. Think Visa. The more, the better.
They clearly see AI token usage as a new commodity and something that will grow and grow for a long time to come.
They want to do the same. Take a small cut from as many tokens being passed around as possible.
Stripe is a financial institution that has strict requirements for reporting and audits by the US government. It has blocked payments for numerous different services for no reason. I wouldn't trust any infrastructure built on OpenRouter.
What exactly is "Open" about OpenRouter anyway?
Echoes of WhatsApp. Huge congrats to the team!
Any companies that are bought go through culture crashes, and I am interested in how this plays out.
It is a known brand for model routing and it has developed means to route model on their own and 3rd-party infrastructure at scale and without much issues.
Why Stripe? Who knows. I don't see much synergy but why not? If anything it is a way for Stripe to get into the market with agentic payments which may come to be worth a lot more than $7B.
Isn't the trend increasingly to let an orshestrator agent decide which model to use in terms of problem difficulty vs. size and thinking budget?
Once you fix those parameters there's little to optimize over, no? I mean you want to find the cheapest one across providers but that's not a recurring decision.
Convenience. Paying only one provider, but get access to different models hosted around the world. Need a model for a project where latency matters? EU hosted and ZDR for another customer? Cheapest for personal use?
You get everything under one roof.
In the second case OpenRouter is very useful; in the first case, maybe it’s an area they move into in time - providing and owning/optimising the orchestrator?
Maybe a signal of how rigid the moats/lobbies are on insurance and private equity
Maybe there is some plan to pump Stripe's valuation to over $1TN before an IPO.
Who knows.
[0] https://www.reuters.com/business/finance/stripe-advent-offer...
The reason is I couldn't find a agent which is as good as Codex/Claude to integrate it. If there is a agent can make full use of it by routing different tasks to different LLMs by difficulty. That would be perfect.
Not sure if there is a agent can achieve this, maybe Deepseek harness have such potential.
I was totally confused by what connected the two companies until I realized "tokens" are basically a nascent, and rapidly growing, "currency".
This is among the more hellish futures I can imagine. Our worth as people reduced to how much access we have to compute. Gross.
there are a few assumptions in this future- the biggest one is - we will have to attain a baseline intelligence which is useful for everyone which is not true today. even fable fumbles hard!!
second is that everyone will have to own a medium to use this intelligence (aka robots). like a currency is useless if you can't use it.
to me this is the least daunting scenario. all other scenarios lead to mass slavery.
That's not the biggest assumption.
The biggest assumption I see is that the population will have the baseline understanding of how to operate in this new paradigm, which seems impossible given how useless many people still are with computers
Incidentally, this is also why I think Bitcoin is such a non-starter for most people
People will revolt first should one have to trade physical goods (such as food that we need for daily lives) with computing power.
It only takes a few people to incite.
you are right, there will be revolt definitely. i think computing power will replace a lot of things though.
I check it's model rankings and prices daily. The charts are great. Please never change!
On another note... I'm surprised OpenAI didn't buy them.
I'm even more surprised OpenAI is giving exclusive discounts on openrouter, essentially encouraging the few API customers they had onto a much larger, cheaper and richer marketplace.
If you have some grand conspiracy, you should be able to validate it. Yes, on occasion, OpenAI has offered Discounts on their API.
is the selling point that they just give you access to a unified access in where you get to write each request toward the particular api provider's api spec? if so why not just sign up with the vendor directly?
Why do they always start with this. Every single one of them. You dont even want to read anything after that. Its the same "Blah blah nothing will change you will get the best of both worlds blah blah"
In this particular case, OpenRouter represents a M x N bridging and enhancement layer much like Stripe themselves, and so must address a lot of parallel technical, dealmaking, accounting, and legal challenges.
It may not be obvious on the surface because they seem to be working in such different domain, but they have to address a lot of the same problems in comparable ways and that makes it a pretty darn good fit.
Founders found out that they could finally afford to pay their medical, tax, housing bills if they said yes
The platform also has another value that is not mentioned often. People often mention the devex, which is consumer value. But one of the main things about open router is that they help model providers access a massive distribution channel with smooth demand, as with any aggregator. Similar to the value food delivery apps provide to restaurants. As such, this type of marketplace power has the potential to allow them to charge model providers commissions as well for various services in the future. You can already almost imagine an "auto mode" where they balance improvements to cx (say right sized models for tasks, saving u money) with "ad"/preferential treatment promises tomodel providers. Basically, all the typical middleman stuff.
One of the other companies in this space, ramp, has also started their own router at router.com. See https://x.com/tryramp/status/2090146780512227825
Quoting their rumored leaked investor letter dated Aug 19 26 below:
Zooming out, we see capital and intelligence are becoming the two digital flows undergirding every business. Up until now, every developer has needed a straightforward and reliable way to manage their revenue pipeline, and serving this need gave rise to Stripe. Going forward, however, every developer will also need a straightforward and reliable way to manage their intelligence pipeline. This observation first led us to OpenRouter. OpenRouter has built the world's largest and most trusted token routing engine, supporting all major models and providers, and beloved by its customers....We think that there are deeper reasons to pursue integration besides convenience. Our experience in working with our customers has led us to realize that intelligence is special: it is expensive, heterogeneous, and constantly changing. As with financial capital, businesses must reason about cost and return of every unit in a deliberate and granular way. How valuable is this task? With which models can it be best handled? Who will pay, and when, and what is the time-value of that delay? We have seen the parallels between managing intelligence and managing capital directly in our own products. Radar, for example, was initially designed to prevent financial fraud, but is proving extremely effective at guarding against token fraud at many of the world's largest A1 companies. Metronome (used by Anthropic, Nvidia, and other industry leaders) is showing that metered billing in an AI context is inseparable from token serving and consumption itself.
They also have very complementary problems and skills. The long-term roadmap for routers is auto-routing; that is, turning traffic into signals into automated decisions. Stripe has a lot of overlapping talent and experience from fraud-detection and likely other products.
What a time to be alive. I’m going to take a 1% stake in a bridge and rename it bridge.ai and sell it for billions too
What could go wrong
I've built one in Rust and want to market it.
So if it’s a network router or a woodworking router, then you’re good.
I can sells of red socks, and white socks, without getting sued by Chicago or Boston.
If it’s in the same space… you might want to contact a lawyer. I believe Firefox had conflicts with earlier names like Phoenix and FireBird and they weren’t even browsers.
Just because a name has open in it doesn't mean it has to be open source.
50x revenue is also a crazy number. I wonder what happens more, companies selling for ≤5% revenue or companies selling for ≥50x revenue.
why is OpenRouter worth 7 billion dollars?
It's a pretty slick system and if you're building on openrouter today and are worried about stripe integration messing up the product, this is a good option to consider.
Separately, congrats to the OpenRouter team. They are selling at the right time in what is undoubtably going to become a contentious market.
Heard it's too hard to implement with Stripe.
[1] https://www.tumblr.com/ourincrediblejourney/89180616013/what...
Are you perhaps thinking of Microsoft acquiring Skype for $8.5 billion? (a similar amount as the OpenRouter deal)
"that will remain essential in a post-AGI economy." — ooof. The magical future economy
In five years they will be struggling.
In ten they will be dead.
holy crap, save some butter for the bread omg
He says it like it’s a huge accomplishment that he achieved when it’s an opinion about something trivial
OpenRouter is a name.
I can delete my comment, though.
AND, if you are too harsh on the former, you incentivize the latter.
Not really. If someone is interested in an LLM's opinion, they can ask it themselves, all LLM providers are one browser tab away.
"I recall topic X being related to fact Y.
Chatgpt says this when asked about how Y relates to X
"blablabla"
"
Hard to think of any type of company that can do the topline meme as effectively as a compute aggregator proxy. Almost like a bitcoin pool mining company counting their shared miners.
Really? 100M annual? That HAS to be profit, it can't be this crazy.
EDIT: From their website, they say they have 200T monthly tokens, at 10 cents per M token ( optimistic average between free tokens and expensive tokens) that's 20M$/month, which is roughly 240M$.
Bonus track: here's chatgpt completely fumbling the above very simple calculation
https://chatgpt.com/share/6a85f831-9118-83e9-8a06-d32195c557...
I don't doubt the chinese bootleg product that openrouter is peddling is even worse than that.
We are in a bubble confirmed.
Sorry for the negativity, but this is highly bearish news.
Lock-in to what exactly? A string(string) function without any further restrictions? Note that aggregator users will not use more specific parameter features (because those vary by vendor), or they escape the aggregator and lock-in directly string(string, vendorOptions={"openAI.logprobs":true})
Selling vendor lock-in to vendor lock-in avoidant customers sounds like a losing proposition, like trying to beat competition on price and undercutting, which is a thing openrouter does too btw, the main reason people use these products is to get cheaper prices. I don't think 2$/M tokens is expensive, and I don't think those that try to cut costs are going to win, whether devs cutting token costs, or small business owners with 5 employees who look at AI as a way to fire 2 of them. I get that the other end of the spectrum is overspending and tokenmaxxing, but the conclusion is that extremes are bad? Nothing new here.