This is a cool way of approaching AI models. I'm a big fan of local LLMs, local specific models like this should be even more powerful.
> Every model is free up to 100k monthly active devices. No tokens, no logins.
I dunno about the business model though. Cloud LLM billing makes sense: you're getting another computer to do work with each request, and using their compute via their gateway that they bill you.
These local models are like old school software. They're producing the weights, and then giving them to people. If I'm happy with the weights you've given me, and I'm not using your compute for inference and not wanting or needing any updates off you, why should you continue getting money off me and my customers?
The whole "but we need to keep it updated for your security" doesn't really work as well for software designed to run fully offline like these local models are.
I'm not saying they shouldn't get paid, but I guess I feel a personal sadness that it's less obvious how to successfully monetise such a sincerely useful and beneficial approach towards AI models.
> If I'm happy with the weights you've given me, and I'm not using your compute for inference and not wanting or needing any updates off you, why should you continue getting money off me and my customers?
I definitely understand the appeal of the desire to “buy it once”, but I think there are a few issues:
- almost no software is static. Look at a package like python Requests and even though it does the simplest thing and has barely changed from a user perspective, it gets updated all the time. This is true for most software. This is doubly true for something like local AI models where both the software and the hardware are changing constantly. Subscriptions motivate sellers to keep their software up to date.
- If I’m an app developer, the idea that I can try something out for $X/month is very appealing versus making an upfront investment of (let’s say) $X*20. This is doubly true for something like local models where I will almost certainly want something new when the models improve.
- To add to the first point, I work at a startup. No one asks questions when I want $20/month licenses. But let’s say I want something that’s gonna be in the 5 figure range annually. If I go to my CFO and ask for $50k upfront and then we implement something and the project fails, I look like an idiot. If I ask for $2000/mo budget for something and then we try it for two months and it fails, no one cares. Subscriptions are just safer in this sense.
I think the issues you're describing are all modern internet software issues. I'd guess you're possibly a bit younger than me, or maybe forgotten how software worked before about 2010.
- All compiled software is static. New versions are released. I'm happy to pay for new versions/features if I want/need them. But if the version I have is working offline, it doesn't need security updates, and if it does what I need, it doesn't need new features. Same applies if I'm building software for others, if the library does what I need for my software for my customers and I need nothing more off the dev, why should I keep paying the dev?
- I'm struggling not be sarcastic here, sorry, but this might shock you to hear: most humans find the idea of being able to try whatever they want and get paid for it appealing. That's not really how money or life works though.
- I don't know how to respond to this point, it's a bit of a strawman. Yes, if the only other option is to pay $50k up front for a local model, then your point is valid.
As a kid, I spent literally hundreds of £ on the first Sims, and Sims expansion packs at £20 /pop. EA got loads of money off me, I got loads of new features I wanted. I moaned at the time, but on reflection, everyone was a winner. Sims is still going now (possibly as a subscription!)
Again, apologies for a bit of sass here (that's a me problem), but the number of people who seem to sincerely believe the points you've put forward is why I'm personally sad at the state of modern software economics.
Very much like map CD-rom updates, I think a lot of people wouldn't mind paying per-update for these models, and would understand the implications. That is, if there weren't so many open models available.
How would a model where you pay per each model/weight update feel like? More similar to traditional software, where you'd get the current version, maybe even free/rebated version upgrade to the next, but then have to do another purchase to get the latest version? Or something similar to that?
Feels like it'd be more aligned with how the business and users would produce and use it, it'd be something like: "We've done a bunch of research and output, here is the thing, you can buy it, and we'll work on the next, then you can pay again for the next if you'd like".
This seems like common sense. I guess it's not seen as common sense any more because the assumption is that all software is going to be using someone else's compute, or be connected to the internet and therefore require regular updates.
If neither of those are true, a subscription or recurring license or similar regular fee just feels like greed/robbery.
How is it not obvious and fair that they are asking you to pay them when you see success (defined as >100K MAU)? How more aligned can you and them be besides this?
When I buy I chair I don't pay a share of my income to the furniture shop when I get rich. I would buy another chair at some point too, maybe fancier one (or the same). Because chairs are commodity. I do however pay taxes to the government based on my income because it keeps doing ongoing maintenance on everything.
Everyone wants to be paid forever for something they produced once is some kind of a mind virus. Make me a better chair and maybe I will buy it, but don't expect to become a trillion dollar company. It's deeply unfair to everyone who wants to be a trillion dollar company of course.
What is really funny to me -- the ones that do make it to collect the rent indefinitely also decrease their own taxes paid to the government AND also decrease the amount of contribution to society by making less and shittier stuff.
> If I'm happy with the weights you've given me, and I'm not using your compute for inference and not wanting or needing any updates off you, why should you continue getting money off me and my customers?
Because they own the IP and they get to decide the terms of how it’s licensed.
This is like a company taking open source software, saying they like the code the community has given them, and asking why they should continue having to respect the terms of the license after downloading the code. The availability of the software (or models) does not equal a free license to use as you please.
A license allowing 100K devices for free is very generous. The businesses selling more than 100K units of anything will be significant operations. It’s fair that they’re asked to contribute financially.
No, you’re arguing against the first part of my comment without reading the whole thing where I addressed the value proposition.
A 100K free device license is a generous gift to small companies. Past 100K, it’s more than fair to ask that companies contribute. That’s a significant operation at that point and the model has obviously provided value.
If the model doesn’t provide any value to a >100K device company then they should train their own or not use it.
I could see how people unfamiliar with per-device or per-unit licensing fees would be confused, but this is a common business model. Scaling payments with the customer’s business success is one of the more fair ways to align vendor and customer while also providing a nice way to give freely to smaller businesses like this.
It’s a good thing. It’s silly that a company giving away a license for 100K devices, which covers small and medium hardware operations, is being criticized for this of all things.
> why should you continue getting money off me and my customers?
Welcome to the concept of fair market value. Less snarkily you're conflating the concepts of price and cost; theyre not the same thing and theyre not the same for you or the seller.
I definitely think there's a lot to be done with small models dedicated to specific tasks. I've always thought the REAL value is in having large models be able to easily build small models for custom tasks (which I know is kind of a thing), but perhaps just providing the small models directly is the more accessible approach.
>accessible via one SDK for Swift, Kotlin, and JavaScript
Lol well let me know when there's a Python SDK and I'll give it a try then. Obviously this isn't a deal breaker if you have a real case, but as someone who is willing to spin something up and try it out if there's a quick "pip install" command, this is getting put back on the shelf for now.
The founders and website seems to talk a lot about mobile devices and how many are being sold/shipped, that's the "unused compute power" they're targeting. So obviously mobile-first platforms and SDKs seem to be the focus first, hence those languages. Understandable, given their target, but just like you I wish there was an easier way to give it a try on a desktop computer.
I would love to use Voz and Ear, but I’d need a version that is competitive with other audio transcription LLMs for platform availability - meaning macOS, Windows and Linux, and supporting GPUs if available.
I love this idea and hope to see more on-device models. How do they make money, though?
I tried out their demo for Clear, the audio quality improvement model.[0] I'm not sure if it's just I don't have refined enough an ear or their demo is broken, but the "raw" and "enhanced" versions sounded exactly the same to me.
The underlying model (DFNet3) is not particularly great but it is very small and fast - imo the best commercially usable denoising model is MossFormer2 (no affiliation - it's just excellent) with one drawback in that it can't remove reverb.
Nvidia's RE-USE model can do what MossFormer2 does _and_ can remove reverb, but it is non-commercial licensed.
at first i got very excited about a new fast transcription model (voz) but turns out its just parakeet v3 with some new inference code which is macOS/iOS specific
It's an ANE optimized version of Parakeet, with our own inference, which enabled us to push performance to about 300x realtime speed on an iPhone 16/17. Our next gen Voz model is trained from scratch and will be at least twice as fast. Android and other platforms will land soon.
A lot of the models would be useful in a web context, to improve on the CMS we're making for clients. But they look like most of them are iOS only, few have a node package or something other, and all the benchmark are running it on modern iPhones so I doubt it would be that fast on a 20$ VPS.
I don't think you understood. It means these are specialized models.
Their toxic model could be ideal for video game lobbies without investing a ton of money if you're an indie dev
This also could be ideal if you want your child to play online to have auto-censorship
It is now also reasonably straightforward if you have access to frontier LLMs, a recent-ish mac and a recent-ish iPhone to point them at the job of porting a given model to run on the ANE - it's a reasonably easy task to hill-climb at this point!
This is a cool idea. The most useful one for me would be something that can process pdf files into a json schema. Title and tag generation from a post would also be useful. I'm interested in web app though.
Once image support drops this could be a cool addition to https://struktur.sh. Will it be on OpenRouter too or will all inference have to be self-managed? I’m thinking about server-side use cases, where budgets are low, so small models shine.
edit: ah I saw mainly iOS for now. But an integration would be possible on macOS then, right?
We plan to make most our models available for Android and web too. Some are a bit harder to port to the different platforms and will take a bit longer to properly land on Android or web. Mostly sequencing (Voz, Clips, Title). Soon!
We design in Figma. Build a design system out that with components for all our public facing products, demos and marketing assets. The website is built with the components, and we have a set of skills and tools to keep both in sync. We also use skills to sync between HuggingFace, our SDK on GitHub and the website. We write and review most our core copy by hand and use that to help expand into different pages. We prob do a write-up about this on our blog later.
This is a cool way of approaching AI models. I'm a big fan of local LLMs, local specific models like this should be even more powerful.
> Every model is free up to 100k monthly active devices. No tokens, no logins.
I dunno about the business model though. Cloud LLM billing makes sense: you're getting another computer to do work with each request, and using their compute via their gateway that they bill you.
These local models are like old school software. They're producing the weights, and then giving them to people. If I'm happy with the weights you've given me, and I'm not using your compute for inference and not wanting or needing any updates off you, why should you continue getting money off me and my customers?
The whole "but we need to keep it updated for your security" doesn't really work as well for software designed to run fully offline like these local models are.
I'm not saying they shouldn't get paid, but I guess I feel a personal sadness that it's less obvious how to successfully monetise such a sincerely useful and beneficial approach towards AI models.
> If I'm happy with the weights you've given me, and I'm not using your compute for inference and not wanting or needing any updates off you, why should you continue getting money off me and my customers?
I definitely understand the appeal of the desire to “buy it once”, but I think there are a few issues:
- almost no software is static. Look at a package like python Requests and even though it does the simplest thing and has barely changed from a user perspective, it gets updated all the time. This is true for most software. This is doubly true for something like local AI models where both the software and the hardware are changing constantly. Subscriptions motivate sellers to keep their software up to date.
- If I’m an app developer, the idea that I can try something out for $X/month is very appealing versus making an upfront investment of (let’s say) $X*20. This is doubly true for something like local models where I will almost certainly want something new when the models improve.
- To add to the first point, I work at a startup. No one asks questions when I want $20/month licenses. But let’s say I want something that’s gonna be in the 5 figure range annually. If I go to my CFO and ask for $50k upfront and then we implement something and the project fails, I look like an idiot. If I ask for $2000/mo budget for something and then we try it for two months and it fails, no one cares. Subscriptions are just safer in this sense.
I think the issues you're describing are all modern internet software issues. I'd guess you're possibly a bit younger than me, or maybe forgotten how software worked before about 2010.
- All compiled software is static. New versions are released. I'm happy to pay for new versions/features if I want/need them. But if the version I have is working offline, it doesn't need security updates, and if it does what I need, it doesn't need new features. Same applies if I'm building software for others, if the library does what I need for my software for my customers and I need nothing more off the dev, why should I keep paying the dev?
- I'm struggling not be sarcastic here, sorry, but this might shock you to hear: most humans find the idea of being able to try whatever they want and get paid for it appealing. That's not really how money or life works though.
- I don't know how to respond to this point, it's a bit of a strawman. Yes, if the only other option is to pay $50k up front for a local model, then your point is valid.
As a kid, I spent literally hundreds of £ on the first Sims, and Sims expansion packs at £20 /pop. EA got loads of money off me, I got loads of new features I wanted. I moaned at the time, but on reflection, everyone was a winner. Sims is still going now (possibly as a subscription!)
Again, apologies for a bit of sass here (that's a me problem), but the number of people who seem to sincerely believe the points you've put forward is why I'm personally sad at the state of modern software economics.
LLMs are static though.
In the way that, say, Garmin Maps are static, I agree.
Very much like map CD-rom updates, I think a lot of people wouldn't mind paying per-update for these models, and would understand the implications. That is, if there weren't so many open models available.
... So we both agree.
How would a model where you pay per each model/weight update feel like? More similar to traditional software, where you'd get the current version, maybe even free/rebated version upgrade to the next, but then have to do another purchase to get the latest version? Or something similar to that?
Feels like it'd be more aligned with how the business and users would produce and use it, it'd be something like: "We've done a bunch of research and output, here is the thing, you can buy it, and we'll work on the next, then you can pay again for the next if you'd like".
This seems like common sense. I guess it's not seen as common sense any more because the assumption is that all software is going to be using someone else's compute, or be connected to the internet and therefore require regular updates.
If neither of those are true, a subscription or recurring license or similar regular fee just feels like greed/robbery.
How is it not obvious and fair that they are asking you to pay them when you see success (defined as >100K MAU)? How more aligned can you and them be besides this?
When I buy I chair I don't pay a share of my income to the furniture shop when I get rich. I would buy another chair at some point too, maybe fancier one (or the same). Because chairs are commodity. I do however pay taxes to the government based on my income because it keeps doing ongoing maintenance on everything.
Everyone wants to be paid forever for something they produced once is some kind of a mind virus. Make me a better chair and maybe I will buy it, but don't expect to become a trillion dollar company. It's deeply unfair to everyone who wants to be a trillion dollar company of course.
What is really funny to me -- the ones that do make it to collect the rent indefinitely also decrease their own taxes paid to the government AND also decrease the amount of contribution to society by making less and shittier stuff.
Analogy doesn't apply to this case, as you are paying for the chair upfront but not paying for the model weights upfront.
I'm also not downloading a chair, because why would it?
Exactly, so your analogy is useless.
> If I'm happy with the weights you've given me, and I'm not using your compute for inference and not wanting or needing any updates off you, why should you continue getting money off me and my customers?
Because they own the IP and they get to decide the terms of how it’s licensed.
This is like a company taking open source software, saying they like the code the community has given them, and asking why they should continue having to respect the terms of the license after downloading the code. The availability of the software (or models) does not equal a free license to use as you please.
A license allowing 100K devices for free is very generous. The businesses selling more than 100K units of anything will be significant operations. It’s fair that they’re asked to contribute financially.
>Because they own the IP and they get to decide the terms of how it’s licensed.
how'd that work out for the rest of IP owners that had their stuff used as training data for use in model creation?
was the material they trained on produced in-house?
Use as training data has been tested in court multiple times and cleared.
I know some people want it to be a legal violation, but it’s not.
Also these are audio models, if you hadn’t noticed. The training sets for this type of work is very different than the corpus of scraped GitHub repos.
Was the material, I was trained, produced in-house?
Philosophical questions don't make sense if you use them to counter argue about something which is a legal issue.
He's not arguing about their right to license it how they please. He's arguing that someone ought to choose to do it another way.
You made up a position to argue against.
No, you’re arguing against the first part of my comment without reading the whole thing where I addressed the value proposition.
A 100K free device license is a generous gift to small companies. Past 100K, it’s more than fair to ask that companies contribute. That’s a significant operation at that point and the model has obviously provided value.
If the model doesn’t provide any value to a >100K device company then they should train their own or not use it.
I could see how people unfamiliar with per-device or per-unit licensing fees would be confused, but this is a common business model. Scaling payments with the customer’s business success is one of the more fair ways to align vendor and customer while also providing a nice way to give freely to smaller businesses like this.
It’s a good thing. It’s silly that a company giving away a license for 100K devices, which covers small and medium hardware operations, is being criticized for this of all things.
> why should you continue getting money off me and my customers?
Welcome to the concept of fair market value. Less snarkily you're conflating the concepts of price and cost; theyre not the same thing and theyre not the same for you or the seller.
How does piracy of digital content and software factor into this concept?
I definitely think there's a lot to be done with small models dedicated to specific tasks. I've always thought the REAL value is in having large models be able to easily build small models for custom tasks (which I know is kind of a thing), but perhaps just providing the small models directly is the more accessible approach.
>accessible via one SDK for Swift, Kotlin, and JavaScript
Lol well let me know when there's a Python SDK and I'll give it a try then. Obviously this isn't a deal breaker if you have a real case, but as someone who is willing to spin something up and try it out if there's a quick "pip install" command, this is getting put back on the shelf for now.
The founders and website seems to talk a lot about mobile devices and how many are being sold/shipped, that's the "unused compute power" they're targeting. So obviously mobile-first platforms and SDKs seem to be the focus first, hence those languages. Understandable, given their target, but just like you I wish there was an easier way to give it a try on a desktop computer.
Same, just wish I could afford the computer to train my own ngl
I would love to use Voz and Ear, but I’d need a version that is competitive with other audio transcription LLMs for platform availability - meaning macOS, Windows and Linux, and supporting GPUs if available.
I love this idea and hope to see more on-device models. How do they make money, though?
I tried out their demo for Clear, the audio quality improvement model.[0] I'm not sure if it's just I don't have refined enough an ear or their demo is broken, but the "raw" and "enhanced" versions sounded exactly the same to me.
[0] https://desertant.com/models/clear/
The underlying model (DFNet3) is not particularly great but it is very small and fast - imo the best commercially usable denoising model is MossFormer2 (no affiliation - it's just excellent) with one drawback in that it can't remove reverb.
Nvidia's RE-USE model can do what MossFormer2 does _and_ can remove reverb, but it is non-commercial licensed.
at first i got very excited about a new fast transcription model (voz) but turns out its just parakeet v3 with some new inference code which is macOS/iOS specific
It's an ANE optimized version of Parakeet, with our own inference, which enabled us to push performance to about 300x realtime speed on an iPhone 16/17. Our next gen Voz model is trained from scratch and will be at least twice as fast. Android and other platforms will land soon.
Nice, now I know what I actually need.
I was also thinking that this is almost too good to be true
A lot of the models would be useful in a web context, to improve on the CMS we're making for clients. But they look like most of them are iOS only, few have a node package or something other, and all the benchmark are running it on modern iPhones so I doubt it would be that fast on a 20$ VPS.
Just a few are iOS first (pure practical timing/sequencing). We plan to make all models available cross-platform in the coming weeks.
> opinionated on-device intelligence
> Hate speech triage. On-device moderation that flags hateful, abusive and threatening text
What could go wrong here?
I don't think you understood. It means these are specialized models. Their toxic model could be ideal for video game lobbies without investing a ton of money if you're an indie dev
This also could be ideal if you want your child to play online to have auto-censorship
I don’t know if i want to expose my child to auto-censorship (or online gaming anyway).
1. Not having human in the loop to review it because humans are expensive.
2. Having human in the loop to review it and subject said human to the worst other humans produce.
Shiny layer of marketing and proprietary code on top of open models?
Voz is Parakeet 0.6B v3
Clear is DeepFilterNet 3
Ear is the language predictor from whisper-tiny
...
It is now also reasonably straightforward if you have access to frontier LLMs, a recent-ish mac and a recent-ish iPhone to point them at the job of porting a given model to run on the ANE - it's a reasonably easy task to hill-climb at this point!
AI beige theme and obnoxious AI writing signaled as much.
seems to be…
This is a cool idea. The most useful one for me would be something that can process pdf files into a json schema. Title and tag generation from a post would also be useful. I'm interested in web app though.
OCR on steroids. Our Schemer model will soon be available (free form text to structured JSON). Once that lands, we want to jump into image to JSON.
But Schemer will be text-only, correct?
Once image support drops this could be a cool addition to https://struktur.sh. Will it be on OpenRouter too or will all inference have to be self-managed? I’m thinking about server-side use cases, where budgets are low, so small models shine.
edit: ah I saw mainly iOS for now. But an integration would be possible on macOS then, right?
I wonder why they only support Apple platforms, citing CoreML. Doesn't Android have a similar framework, ML Kit?
We plan to make most our models available for Android and web too. Some are a bit harder to port to the different platforms and will take a bit longer to properly land on Android or web. Mostly sequencing (Voz, Clips, Title). Soon!
The website looks amazing.
Thanks!
How was it built?
We design in Figma. Build a design system out that with components for all our public facing products, demos and marketing assets. The website is built with the components, and we have a set of skills and tools to keep both in sync. We also use skills to sync between HuggingFace, our SDK on GitHub and the website. We write and review most our core copy by hand and use that to help expand into different pages. We prob do a write-up about this on our blog later.