OpenRouter, Inc.

Meta is pitching it for coding and agentic work, meaning multi-step tasks where the model plans, uses tools, and keeps going without a human at every click.

Hosted copies are already on API platforms, including Together AI, Fireworks, Groq, and OpenRouter, and the files are on Hugging Face plus the big cloud model gardens.

Meta also says two Apple M5 Ultra Mac Studios running Apple's MLX software can hold the whole model in memory and drive it on a desk, which is a first at this class if the claim holds up for your workload.

The Muse model license lets many developers use it free, with limits on some commercial cases, so read it before you ship.

Very cool.

Love this." And then I ran into this issue where I'm like, okay, well, all of this works now, but the problem is is OpenRouter has, [laughs] like, 130 models that you can use.

And I'm trying to, like, brainstorm on reconciling this in the UI, 'cause I'm like, well, I obviously don't want them to, like, open the model picker on the chat and have 130 different options.

And eventually we just decided, like, let's just pick the top five.

And it plugs into other memory systems it was never even trained on and still wins.

How does that even work? OpenRouter launched Armor in beta a day before that.

Screens agent tool calls before execution, claims it beats Claude Code's own classifier at a fraction of the cost.

And over on GitHub, top trending repo is called Rhea.

So Venice combines a lot of different things in one package.

You know, it's kind of like OpenRouter in the sense that you can choose the model you want, you can use their default model, you can use open weight models, or you can even use frontier lab models like like ChatGPT or Claude.

Those are all available under Venice AI.

And the key distinction is privacy.

The stuff they are seeing in terms of the way Russians are using, uh, agents now to hit targets in Ukraine, uh, it's causing real problems for the Ukrainians, right? Because they are just able to scale up and do everything at once.

The crazy thing is, though, you, you could actually see, if you're doing incident response across a whole bunch of these different incidents, you could see where the Russians ran out of tokens via OpenRouter, 'cause everything simultaneously on 10 different targets just stops.

It's dead for three hours, and then someone, you know, plugs a credit card number or whatever into OpenRouter, and bang, um, up it goes again and, and it all starts.

So, you know, there's a lot happening at the moment in terms of, um, real harms, not these theoretical self-replicating science fiction harms.

And, you know, the, the, the public discourse around this stuff on, like, X, for example, is just so bizarre.

biggest single-day surge in the whole trending pool.

Meanwhile, on OpenRouter, the model rankings are swinging wildly.

DeepSeek's leading at 37.43 trillion tokens, while other models dropped 61% in the same window.

But the app leaderboard underneath all that? Didn't move.

If he wants to build something on DeepSeek, he has four routes.

The official API, OpenRouter, a specialist host, or renting a GPU and serving the weights himself.

Which one, and why?

That last part is the one that actually bites.

That cache number is the one to remember.

It is.

Now OpenRouter reselling the same model is fifteen cents in and sixty cents out.

Deep Infra is nine cents in and eighteen cents out.

Whereas it used to take, hey, somebody released something, you'd have a two-year window, three-year window to exploit it before somebody came running next to you with your com- better, slightly better competitor, and then you had to fight it out for market share.

So talk about on a business level what it's like to run this company specifically because you have NVIDIA, who I assume is a partner, but maybe with Open Router and some of the things they're doing competitive, but they're also making Nemotron and they bought, what was the, uh, OpenPool or what was the other, uh, Laguna? They bought that I think.

It was Poolside.

Poolside.

Uh, does that resonate with you at all? Do you hear that from your customers, the, this like resiliency and longevity of the platform?

Definitely.

And, and, you know, uh, just to jump back to NVIDIA for a minute, you know, w- they're a great partner of ours, and whether it's the Nemotron models, the Open Shell project, which solves this kind of problem directly, um, which is how do you bring, um, security and dependability and reliability to a world where everything's changing? Um, and so I think we're seeing a new breed of software, and I think, you know, Open Router was a great example of a really fast ascent and, and example pr- product where you ask a customer, um, using Ollama, using a tool like Open Router, why, and it's largely they're by or fall in the generative media space, is that they're, they're able to keep control of the workflow, but-

Mm