Own It vs Rent It - The West's Answer to China's Open Models (and a Top 30 pat on the back).
With Jez Rowe and Cam Lynch
In October 2026, New York startup Reflection AI released Beam and France's Mistral followed a day later with a trillion-parameter model nicknamed Le Chonk: the West's answer to China's open source AI. Jez and Cam see it as a real win for enterprises and for countries like Australia without a frontier lab. It isn't a small business play, though. For SMEs, renting cheap models through an API still beats owning the hardware.
What this episode answers
What is Reflection AI's Beam open source model?
Beam is an open source AI model from Reflection AI, a two-year-old New York startup, unveiled in October 2026. Anyone can download it, run it on their own computers and change it, unlike ChatGPT or Claude, which you rent. It has 501 billion parameters but only 23 billion switch on for any one job, like a 500-person staff where only 23 go to each meeting, which makes it cheap to run. A day later, France's Mistral released a trillion-parameter open model nicknamed Le Chonk.
What is the difference between open source and closed AI models?
An open source AI model is one you can download, run on your own hardware and modify. A closed model, like ChatGPT or Claude, is one you rent through an app or API. Open models give you control and keep your data in house, which is why enterprises are building on them. Closed frontier models are still the smartest, and they stay expensive. The emerging pattern uses both: frontier models to orchestrate, and cheaper workhorse models to do the grunt work.
Is it cheaper to run an open source AI model yourself or pay for API tokens?
For a small business, paying for API tokens is far cheaper than running an open source model yourself. Cam Lynch, founder of Vibe CFO, has tested it: a Mac Studio with about 256GB of RAM, costing around $15,000, can run a 300-billion-parameter open model, but it's slow and only one person can use it at a time. In October 2026, OpenAI's cheap workhorse model cost around 10 cents per million tokens through the API. Enterprises building proprietary models on their own data are a different spend entirely.
Why are big companies building their own AI models instead of using OpenAI or Anthropic?
Big companies are building their own AI models because their data is their edge, and they don't want to hand it to a frontier lab. The labs have the smartest models, but enterprises hold something the labs don't: their own data, which is what makes AI useful in their domain. Trust took a hit with the Claude Design and Figma story, which reportedly ended with Figma closing its partnership with Anthropic. Many enterprises responded by building on open source models, accepting a little less capability for control.
Why do Western open source AI models matter for Australia?
Western open source AI models give businesses a capable open option that isn't Chinese, and bring AI control onshore for countries like Australia that don't have a frontier lab. Chinese labs led open source AI, but Western enterprises wanted an alternative they trusted on security. As of October 2026, the new American open model sat roughly one step, or three to six months, behind the best Chinese ones, which is good enough. As Reflection AI's Misha Laskin put it, the only way to own intelligence is if it's open.
Episode notes
Enterprises want AI built on their own data, and they don’t want to hand that data to someone else to do it. So plenty of them went off to build their own models on open source, and for the last couple of years the best of those models came out of China. This week the West stepped up. A New York startup called Reflection unveiled Beam, and France’s Mistral followed a day later with a trillion parameter model they’ve nicknamed Le Chonk.
What we land on is that the pitch isn’t “we’re the best”. It’s “we’re good enough, and built in the West”. Sitting a step behind the leading open models is close enough to give Western businesses the confidence and security to own their AI rather than rent it.
That’s the real fight now, open versus closed. The smartest AI stays expensive, the workhorse models do the grunt work, and the closed labs are answering with cheap workhorses of their own.
For small and medium business, though, owning isn’t the play. A top-spec desktop box costs around fifteen grand, runs slowly and serves one person, when the same kind of model is available for cents. Enterprise builds its own. Small business gets the benefit without the build. And for Australia, with no frontier lab of our own, open models are how some of that control comes back onshore. As Reflection’s founder put it, the only way to truly own intelligence is if it’s open.
Jez and Cam on the West’s answer in the open model race, and how small business gets the upside of all this without carrying the cost of owning it.
Five minutes. One thing that matters. We ignore the rest, so you can too.
Vibe CFO is Australian founded. An AI CFO for small and medium business, out of the box, governed and secure.
Vibe in 5.
Transcript
Read the full transcript
Ah, yet another big week in AI. This is The AI Exhale, hosted by Vibe CFO with Jez and Cam. Five minutes, one thing that matters, or maybe two. We ignore the rest so you can too.
Jez: Well, I might actually start the episode with a little bit of a pat on the back for both of us, mate. We hit the top 30 on accounting podcasts.
Cam: Oh yeah, booyah!
Jez: So there we go, we're indexed on that. Plenty of good podcasts on there to check out as well. All right, we're going to talk today about the West going open, in regards to open source.
So on Monday, Reflection AI, a two-year-old New York startup, unveiled Beam. It's an open source model. Anyone can download it, run it on their own computers and change it, unlike ChatGPT or Claude, which you rent. This one has 501 billion parameters, but only 23 billion switched on for any job. So it's like having a staff of, say, 500 employees, but only 23 go to the meeting, right? And it's cheap to run.
That's the main point I'd say here today.
And then only a day later, French lab Mistral, which we've talked about, unveiled a large one as well, with a trillion parameters. And I love this one, they nicknamed it Le Chonk. So what does this mean, Cam? I mean, this is the West. We've taken a little bit of a pullback from China, with all of their open source models, which obviously undercut on cost.
Cam: This is a story we've been telling over our podcast, where initially the frontier models came in and they went into enterprise, and they wanted to basically give full-blown AI to these guys in enterprise.
But then the Claude Design and Figma story massively rattled them. Anthropic created Claude Design, and they had a big bust up over it. And Figma closed down their partnership with them. And the story that's been told is that Anthropic was taking their data. So then all the enterprise said, stuff this, we're going open source. We'll build our own models. They won't be as good.
But they have something that the big frontier labs don't have: the data, their data.
Which is crucial, because if you want to use an AI, say take Figma with how good it is at design, it's crucial that you have the designers' data and what they're doing to make the AI useful. This was the big movement, right, where the Chinese labs came in, and they weren't as good as the frontier, but they were pretty good. Remember our Kimi K3 episode? That was one of the first that came out.
The Chinese labs have got really, really good, and this is where the enterprise has gone off trying to build themselves. But the pressure has mounted on the West, that the West needs an open model, because all of a sudden we've got all these big enterprises building. So we needed an American or Western open source model. And so the day has come.
Jez: And so I guess with this open source model being out of the US, as you speak to the geopolitics, their pitch isn't that we're the best, it's just that they're not Chinese. And that's kind of what matters around, I guess, security in the West.
Cam: I think it's the security thing, but it's also good enough, right? The best Chinese open models are around GLM 5.3, and this American model is roughly at GLM 5.2 level, one step behind. To sit three to six months behind the Chinese open models is pretty good, and it just gives you confidence. American businesses will have confidence that the route forward is to use open American models. And then OpenAI brought out their cheap model at 10 cents per million tokens.
So they've got workhorse AI as well. That's sort of their response to this. They say, look, use the frontier for orchestration and then use our workhorses, and you don't need to build your own model. So the battle's going on between open and closed right now.
Jez: Yeah. So I guess the smartest AI stays expensive, and this AI will be doing most of the work, sort of that grunt work.
Cam: Like, when we look at this, I don't think that this is in the realm of small business. If you take my Mac Studio, right, which has got about 256 gigs of RAM, we can run the GLM 5.2 Flash, which has 300 billion parameters, right? But it's pretty slow. It's not great. And it's only one person who can use it on the Mac Studio. And the Mac Studio costs what, $15 grand? So it's not really that economical.
Compared to paying 10 cents per million tokens directly through the API, you're sort of like, this is crazy. For a small business, it's crazy. For an enterprise, they're thinking in their head, no, no, we're going to build our own model and it's our proprietary model. So it's a different spend for them compared to us in small to medium business land.
Jez: Yeah, but it does bring this sort of power to Australia, if we're speaking closer to home.
We don't have our own frontier lab. So this open models piece does, you know, bring it onshore, to where we actually have a feeling of control.
Cam: Definitely. And it's getting to that stage where it's getting better and better.
Jez: So yeah, well, that guy out of Reflection, the New York startup, Laskin, said the only way to own intelligence is, by definition, if it's open.
And that's your five-ish minutes back. Vibe CFO, Australian founded. An AI CFO for small and medium businesses, out of the box, governed and secure. vibecfo.ai. Now you know.
Now you know.
The AI Exhale is made by VibeCFO. Australian founded, an AI CFO for small and medium business, out of the box.