Re:Qwen4-Max
by rbehrends on Monday November 16, 2026 @06:56PM
(#50525372)
I think the jury is still out on this. Other companies are instead focusing on fine-tunes.
There are a number of issues with relying on a frontier LLM for these kinds of tasks.
One, there are only 3-5 companies in the world that have the capability, and you depend on whether they're interested enough in your use case (do you think you can interest Meta in putting support behind optimizing support for Swedish law, for example?). These companies measure their investments by ROI, and they focus on things that have large markets. Heck, Google in particular is kind of infamous for suddenly killing off projects that don't work for them.
Second, all of these companies are American, which may be a no-go for certain use cases involving strict data privacy.
Third, these capabilities may need to be maintained and updated, and you'd depend on the goodwill of frontier providers to keep doing that.
Fourth, the frontier model is likely to be more expensive to deploy and run than a smaller specialized model.
You may have a point in a world with infinite resources, but the reality is that this situation is influenced by plenty of market constrains.
There are a number of issues with relying on a frontier LLM for these kinds of tasks.
One, there are only 3-5 companies in the world that have the capability, and you depend on whether they're interested enough in your use case (do you think you can interest Meta in putting support behind optimizing support for Swedish law, for example?). These companies measure their investments by ROI, and they focus on things that have large markets. Heck, Google in particular is kind of infamous for suddenly killing off projects that don't work for them.
Second, all of these companies are American, which may be a no-go for certain use cases involving strict data privacy.
Third, these capabilities may need to be maintained and updated, and you'd depend on the goodwill of frontier providers to keep doing that.
Fourth, the frontier model is likely to be more expensive to deploy and run than a smaller specialized model.
You may have a point in a world with infinite resources, but the reality is that this situation is influenced by plenty of market constrains.