Original Title: Owning vs. Renting Intelligence
Original Author: Lin Qiao
Compiled by: Peggy, BlockBeats

Editor’s Note: Mythos was shut down this week, reminding many AI entrepreneurs of an issue obscured by cost discussions: when a product’s core capability is built on external models and platforms, what does the company truly own?

In recent years, open-source models have often been discussed within the framework of “cheaper frontier model substitutes.” But this article argues that cost is not the most critical variable—control is. For an AI company, calling a frontier model API can quickly launch a product and lower technical barriers, but it also means that core capabilities may be constrained by the model provider’s rules, pricing, strategy changes, or even takedown decisions.

The article further proposes that “owning intelligence” does not mean giving up frontier models; rather, businesses should accumulate their own data, workflows, domain knowledge, evaluation standards, and edge cases into a controllable model system. The future competition in AI may not be dominated by a single largest model; instead, multiple “frontiers” will emerge: general frontier models, enterprise-specific post-training models, vertical specialized models, and routing systems composed of multiple models working together.

Mythos’s shutdown, therefore, feels like a reminder: in the AI era, the real moat is not just whether you can call very powerful models, but whether you can turn intelligence into assets of your own company.

The following is the original text:

Mythos was shut down this week. Whether you agree with this decision or not, that’s no longer the point.

What really stings many people is this: a company built on intelligence it cannot control suddenly finds itself under a set of decisions it cannot influence. After seeing this, many founders ask themselves the same question: which parts of my business are actually just “rented”?

In the past few years, discussions about open-source models have mostly focused on cost: can they really accomplish tasks? If they can, how much cheaper are they compared with calling frontier model APIs?

By now, we already have a fairly clear answer. We’ve worked with companies like @RampLabs, @cursor_ai, @harvey, and so on, and the basic path is similar: start from a powerful open-source model, post-train it on the work that is truly important to the company, and continuously compare it with frontier models through rigorous evaluation.

The results repeatedly come as a surprise. For the tasks that enterprises care about most, a tuned open-source model can often get close to—and even reach—the quality of frontier models at a very low cost.

But what truly became clear this week is that cost has never been the most important issue.

The deeper problem is control. Who actually owns the intelligence your product depends on?

Recently, many discussions have been summarized as the difference between “renting” and “owning.” This analogy isn’t perfect, but it’s useful.

Renting Intelligence

Before things go wrong, renting has always worked well. Apartments are move-in ready—you can carry in your belongings and live there; the lights can turn on, the plumbing works, and maintenance is handled by someone else. That’s why most companies choose this path at first.

Frontier model APIs are excellent products. They enable startups to build things that, a few years ago, would have seemed unimaginable.

But renting also comes with limitations. Landlords can raise rent, decide what kinds of renovations you’re allowed to make, and change the rules. Occasionally, for reasons that have nothing to do with you, they can even tell you to move out.

You haven’t done anything wrong. You’re just operating on someone else’s territory.

That’s also why Mythos’s story resonates with so many people. When your core capabilities depend entirely on someone else’s platform, you expose yourself to a set of decisions you can’t control.

Most of the time, this isn’t important. But sometimes, in an instant, it can become extremely important.

Owning Intelligence

The lesson here isn’t that companies should stop using frontier models. Far from it. Frontier model labs have achieved extraordinary technology. Most products should use them. We use them ourselves.

In many ways, frontier models are becoming infrastructure. But infrastructure and ownership are two different things.

You can use public infrastructure while still owning what truly creates value for your business. In the AI field, what people mean by “owning” is starting from a state-of-the-art open-source model and shaping it around the most unique parts of your company.

Your data.

Your workflows.

Your domain knowledge.

Your edge cases.

Your evaluation standards.

Your definition of “good.”

Over time, this model will become less general and more capable of reflecting the work your company actually handles every day. Value is created right here.

You can think of it like a house. Moving furniture is easy, and repainting a wall is easy. But if your future depends on the layout of the house itself, you’ll sooner or later want the ability to move walls. Intelligence is the same way.

When intelligence truly belongs to you, no one can quietly pull the floor out from under your product.

That’s also why we built Fireworks this way.

We place training and inference within the same system, allowing companies to adopt the best open-source models, shape them around the most important problems in their own business, and deploy them stably into production.

Not just consuming intelligence. It’s owning intelligence.

No single frontier

There’s another optimistic takeaway this week: the future of AI doesn’t depend on one model winning everything.

There is no single frontier. There are many kinds of frontiers.

A frontier model is one kind of frontier.

A model fine-tuned based on years of a company’s proprietary knowledge is another kind of frontier.

A specialized model that solves a narrow problem better than any other model is yet another kind of frontier.

A system that routes requests to multiple models and allows them to collaborate, surpassing single models across many tasks, is also a frontier.

The most interesting change in the AI field isn’t that one model is getting smarter and smarter—it’s that intelligence is becoming more and more customizable.

In the end, the companies that win aren’t necessarily those that have the largest models, but those that can turn intelligence into their own unique assets.

Looking Ahead

This week, a lot of time was spent reacting to news, but we chose to keep releasing products: @Kimi_Moonshot K2.7 Code, @MiniMax_AI M3, @Alibaba_Qwen 3.7 Plus.

The future I’m looking forward to isn’t one where a model quietly swallows everything it sees.

It’s one where many teams can each own their own portion of the frontier.

If Mythos being shut down makes you start rethinking the trade-offs involved, we’d be happy to talk.

[Original Link]

Click to learn about Rhythm BlockBeats job openings

Welcome to join the official Rhythm BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Twitter Official Account: https://twitter.com/BlockBeatsAsia