What does dual-wielding AI mean?

Using two AI tools for different jobs — typically a premium model for high-stakes work and a cheap open-source model for everyday tasks. It saves money but requires you to track which model is best at any moment.

Which AI model is best for everyday tasks?

Capable open-source models handle everyday tasks — emails, planning, questions — at a fraction of the cost of frontier models. The specific best one changes often, which is why a maintained default is easier than a manual choice.

Should I pay for two AI subscriptions?

Usually not. One subscription with automatic routing covers most people. Paying for a premium tool only makes sense if your work genuinely needs frontier capability every day.

Why are cheap AI models good enough now?

Chinese open-source models improved dramatically over the past year and now own the best cost-to-output ratio for low and medium difficulty work. They are strong models that cost pennies to serve.

What is the downside of using cheap Chinese AI models?

Two things: the best model changes constantly, so you must track releases to stay on a good one, and there is no polished consumer app for them — the labs all build for technical users. The models are excellent; the way in is not.

Is Claude better than DeepSeek?

For high-stakes, first-time-right work, yes — the frontier earns its price there. For everyday drafting and questions the difference is small and rarely worth a 4x price gap. It depends on the stakes, not the model ranking.

Why dual-wield AI? Because there's no single best model — and that's the problem

There is no single best AI model, so power users dual-wield — frontier models for high-stakes work, cheap open-source models for everything else. Here's my real split, why it works, and why most people should not copy it.

Is there one best AI model?

No — and the proof is that the people who know AI best carry two subscriptions. There is no single best model because different models win at different things, and which one wins changes every few weeks. "Dual-wielding" AI is the honest workaround for a market that refuses to hand you one right answer. It works. I use it every day. And I will also tell you, plainly, why you probably should not do it.

How I actually dual-wield

Here is my real split:

High-stakes work: Claude, the current frontier pick. When the output has to be right the first time, I use the strongest model available and pay what it costs.
Everyday life and side projects: cheap Chinese open-source models. They are genuinely strong now, and they cost a small fraction of what the Western frontier charges. For drafting, planning, quick questions, and all the low-stakes work that fills a normal week, I cannot tell the difference often enough to justify the premium.
Sometimes others. Gemini for pulling together information scattered across sources, occasionally something else when a new release is making noise. The option set keeps changing, so the split keeps shifting too.

The logic underneath the split is simple: pay for capability where stakes are high, save everywhere else. I get real value from both ends.

The economics behind the habit

Why does this split work at all? Because the two ends of the market have different cost structures. Frontier Western models are expensive to serve — that is where their price comes from, as I break down in why ChatGPT is so expensive. Chinese open-source models cost a fraction to serve, so apps built on them can charge a fraction. Same everyday capability, 75% off the subscription. The dual-wield is just the power user's way of capturing that gap manually: premium where it pays for itself, cheap where cheap is indistinguishable.

But here is the part I will admit freely

Dual-wielding is unnecessary for most people. Two problems come with it, and I feel both of them personally:

You have to pay attention every day. The best cheap model changes constantly. Labs ship updates every few days, leadership rotates, and a model you loved in spring can regress by autumn. Keeping track is a hobby, not a habit worth forming. I happen to enjoy the hobby — most people sensibly will not.
There is no good interface for the cheap models. The open-source labs all target technically adept people. Their apps assume you know what a model is and enjoy picking between them. There is no clean website you can point your parents at. The models are excellent; the way in is not.

So the honest framing: dual-wielding is a hobbyist's answer to a problem an app should have solved. I enjoy it the way some people enjoy tuning their own car. That does not make it sensible transport.

The turn: let an app do the dual-wielding for you

The dual-wield instinct — right tool for the task, premium only when it matters — is correct. The execution is what is broken. Doing it manually means becoming the router yourself: tracking releases, judging stakes per task, maintaining multiple subscriptions, and re-doing all of it every time the frontier moves.

My belief, and the reason I built Plainly, is that a website should do this for you. One subscription. The app keeps testing the frontier and sets the default to the best available model. You just ask. You get the saving of the dual-wield — most of your traffic lands on efficient open-source models — without the daily attention it demands. Most people should use a tool that picks the model for them; that argument in full is which AI model should I use.

The honest limit

If you are technical and enjoy it, dual-wield. It genuinely saves money on low-stakes work and buys capability where it counts — I would not run my own setup any other way. But if you have ever sighed at the words "model selection", the manual version of this is not for you. The maintained-default version is. For what the cheap half of the split costs and what it covers, see is a $5 AI good enough.