Sam: Seven billion dollars? For an API router.
Kai: I'm Kai.
Sam: And I'm Sam.
Kai: And this is Devs You Should Know! It's Thursday, August 20th, 2026.
Sam: And we have a packed show today, all from the world of open source.
Kai: First, OpenRouter, the AI model routing service, is joining Stripe. And the rumored price tag… is staggering.
Sam: Next, we're asking why Google is making it harder to verify Android's source code, and who really gets hurt.
Kai: And finally, a huge new release from Unsloth promises to make local LLMs faster for everyone—and it's free.
Kai: Okay, let's start with that big acquisition. OpenRouter, the LLM API aggregator, just announced they’re joining Stripe.
Sam: 'Joining Stripe.' I love the careful wording. But let's get to the number everyone was talking about before it was official.
Kai: Oh yeah. Seven. Billion. Dollars.
Sam: For what's basically a reverse proxy for LLM APIs. I mean, it's a great tool, it standardizes inputs across models… but seven billion?
Kai: Come on, it's more than a proxy! It's a universal API. It finds the best price-performance, handles retries, fallbacks... It's the plumbing for the entire AI economy.
Sam: And Stripe wants to own the plumbing. It's a classic Stripe move. They're not just a payments company; they're an infrastructure company.
Kai: Exactly! And now AI-powered businesses are just... businesses. Stripe is betting that every company will need an 'AI faucet' and they want to be the ones selling the fixtures.
Sam: But you have to ask, does this kill the 'open' in OpenRouter? Stripe is famously... not open source. What happens to the community? The transparency?
Kai: Well, their post says it'll keep operating as-is, just 'supercharged by Stripe's resources.' I'm optimistic. This could mean more models, better reliability.
Sam: Yeah, 'supercharged' usually means 'folded into a new enterprise SKU later'. So what does this mean for developers today?
Kai: For now? Nothing changes. Except your confidence that the service will actually stick around. If you're building an AI app, it just got a lot less risky.
Sam: Or more risky, depending on how you feel about being a rounding error in a megacorp's roadmap. We'll see what happens to their pricing.
Sam: Alright, next up is something a little inside baseball, but with huge implications. It looks like Google has stopped pushing Git tags for some of its Android source code.
Kai: Okay, I saw this blow up. This comes from the GrapheneOS project, who are... let's just say, meticulous about tracking this stuff.
Sam: Meticulous is one word for it. They noticed that for recent security patches, Google is pushing the code but not the 'tags' that officially label a commit as, say, 'android-14.0.0_r42'.
Kai: So wait, the code is still there, but the official signpost is missing? Why does that matter?
Sam: It matters a lot for verification. Tags are cryptographic pointers. They're how projects like GrapheneOS or CalyxOS prove they're building from the exact source Google claims is the official release.
Kai: Without the tag, they're just pointing at a commit hash that could be... anything? I mean, it's way harder to verify.
Sam: Exactly. They now have to manually hunt for the right commits. It introduces a risk of error or, if you're paranoid, a risk of Google pushing code that isn't what they say it is.
Kai: Whoa. So is this malicious, or just... Google being Google?
Sam: You know the saying: 'never attribute to malice that which is adequately explained by…' whatever. It could just be a process change. Maybe the team responsible got re-org'd into oblivion.
Kai: But the GrapheneOS folks seem to think it's a deliberate move to make their lives harder.
Sam: They do. They see it as one more barrier, making it more difficult for alternative Android projects to provide timely security patches. It really undermines the 'open' part of AOSP.
Kai: So, what's the takeaway for the average Android dev?
Sam: If you're just building a normal app, probably nothing. But if you work on custom ROMs, your job just got harder. It's a reminder that 'open source' isn't the same as 'open collaboration.'
Kai: Alright, let's end on something we can all run on our own machines. Unsloth is back with what they're calling Dynamic 3.0 GGUFs!
Sam: Unsloth. They move so fast. Weren't we just talking about them a few months ago?
Kai: We were! They're the team that made fine-tuning LLMs way faster with way less VRAM. Now they're applying that same magic to inference.
Sam: And specifically to GGUF, which is the format everyone uses for running models on CPUs and Apple Silicon.
Kai: Yep. They claim their new GGUFs can give you up to a 2x speedup and use 20% less memory compared to the standard ones.
Sam: Okay, 'up to' is doing a lot of work there. What's the catch?
Kai: The 'catch' is that it's a new, smarter quantization method. They've combined a few techniques to pack the model's weights more efficiently.
Sam: Is this a new file format? Do I need a special version of llama.cpp to run them?
Kai: That's the clever part: no! They're 100% compatible with the standard GGUF format. Any runner like llama.cpp or Jan can run them out of the box.
Sam: Wait, so they're compatible, but faster? How does that work?
Kai: It's all about the quantization inside the file. They found a way to arrange the bits that's more efficient for the CPU to process, but without breaking the GGUF spec. It's brilliant.
Sam: Okay, that is impressive. So it's not a new engine, it's just a better-packed suitcase. What's this mean if I want to run Llama 3 on my Macbook?
Kai: It means you should look for the Unsloth-quantized version of that model on Hugging Face. You'll get more tokens per second and might be able to run a bigger model. It's just a straight-up win for local AI.
Sam: I'll believe it when I see the benchmarks on my own machine, but I'm cautiously optimistic. A free performance boost with no compatibility hit is rare.
Sam: Alright, let's do a quick recap.
Kai: Stripe's buying OpenRouter, making a massive bet on owning the infrastructure for AI.
Sam: Google is making life harder for custom Android projects by hiding its Git tags.
Kai: And Unsloth is giving everyone a free speed boost for running local language models.
Sam: And before we go... Kai, did you see the top comment on the OpenRouter story?
Kai: The one that just said 'I built an API router on a weekend. Where's my 7 billion dollars?'
Sam: Yep. A good reminder that on Hacker News, every company is worth either zero or a trillion dollars. There is no in-between.
Kai: That's our show! I'm Kai...
Sam: And I'm Sam. We'll be back tomorrow. Don't go starting any seven-billion-dollar weekend projects without us.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.