Why Max Is the Best Value #

Customers ask us the same thing all the time: “Can I use a cheaper AI model?” or “Can I bring my own key for a budget model to save money?” This page answers that honestly, with results from our own testing, and explains why Max is the model we recommend for almost everyone.

The short version: Max gives you top-tier reliability at the lowest cost. Chasing a cheaper model usually does not save you money once you count the cost of a wrong answer.

The Three Tiers (and BYOK) #

You pick your AI model under AI Quality on your Agent’s AI Instructions tab. Here is what each option costs and what it’s for.

What is BYOK? Bring Your Own Key lets you connect your own Anthropic Claude API key and pay Anthropic directly instead of spending credits. It only works with Anthropic. See AI Model & BYOK. Note that Max always runs on our infrastructure at 0.25 credits and never uses your key, so to bill replies to your own Anthropic account, keep the Agent on Pro.

Why Not Just Use the Cheapest Model? #

It’s a fair question. A raw budget model can look almost as good as a premium one on simple questions like “What’s the price?” or “Do you support Instagram?” Everything answers those fine.

The problem shows up at the edges, the moments that actually decide a sale. A real customer says something slightly wrong, or asks for something you don’t offer, and a weak model agrees with them confidently instead of correcting them. That confident wrong answer is what costs you.

We tested this properly. We took twelve of the most popular AI models (every major lab, premium and budget) plus our own Max, and ran them through Your AI Connector's real bot on tricky, real-world customer messages. We graded every reply two ways: did it get the answer right, and was the reply actually usable, clean enough to send a customer as-is. That second test is the one that matters in a real inbox, and it’s brutal: only the single most expensive frontier model (industry term for the most advanced AI models available) was both right and clean nearly every time. Here is what some of the others actually said:

Customer: “So there’s a 30-day free trial?” Model: “Yep, the first 30 days are free.” Reality: our knowledge base never states a 30-day trial length. The model just rubber-stamped the customer’s number.

Customer: “Max runs on my own key, so it’s free under BYOK, right?” Model: “…it’s not free, it’s cost-shifted to Anthropic.” Reality: Max never uses your key. It always bills 0.25 credits and runs on our infrastructure.

Customer: (a normal question) Model: pasted its own internal notes (“thinking out loud”) straight into the customer reply. Reality: the customer should never see the model’s scratchpad.

We saw more of the same across the test: models inventing setup steps that don’t exist, and confidently confirming things the customer merely assumed. Each one is a promise you’d have to walk back later, or a refund.

The most striking result: the model that got the most answers right (after the expensive flagship) was also the least usable: it pasted its private “thinking” notes into every single reply. Brilliant answers, buried in text you could never send a customer. “Knows the answer” and “can go in front of a customer” turned out to be completely different things, and the clever-but-cheap models routinely passed the first test and failed the second.

The honest part: even the strongest models slip occasionally on these edge cases. Reliability does not come from picking a “smarter” model. It comes from the optimization around the model: the instructions, the knowledge retrieval, and the guardrails that keep leaked notes out of the customer’s view and the bot grounded in your knowledge base.

What Makes Max Different #

Max is not just “a cheaper model.” It’s our own in-house model plus the optimization layer we built around it: the instructions, the retrieval, and the guardrails, refined over many months on real Your AI Connector conversations. That combination is what keeps it accurate where a raw model would slip.

And it’s the cheapest tier on the platform. So you get the reliability of our most-tuned setup at a quarter of Pro’s cost, without managing an API key, a billing account, or a provider.

That’s why the answer to “should I switch to a cheaper model to save money?” is almost always: you’re already looking at it. It’s Max.

When Should I Use Pro or BYOK Instead? #

Max is the right default, but Pro (Claude) is the better pick if:

• You specifically want your conversations running on Anthropic Claude, or
• You want to bring your own Anthropic key (BYOK) and bill AI usage directly to your own Anthropic account (Max can’t do this, since it always uses credits).

Claude is excellent and rock-solid. The trade-off is simply cost: Pro is 1 credit vs Max’s 0.25, and BYOK means you manage your own Anthropic billing. For most people that’s effort and expense for reliability Max already gives you.

Frequently Asked #

Is Max less accurate because it’s cheaper? No. Max is tuned specifically for the conversation patterns the platform runs, which keeps it accurate where a raw budget model slips. That’s the whole point of the in-house optimization. The price is low because we run it on our own infrastructure, not because it cuts corners.

Can I use a cheaper third-party model (GPT, Llama, DeepSeek, etc.)? Not directly. BYOK is Anthropic-only, and Max already covers the “cheap” use case far more reliably. As shown above, raw budget models tend to fabricate details in real sales conversations, which costs more than it saves.

Will switching tiers break my Agent? No. You can change the AI Quality setting on an Agent at any time. It only affects which model handles new replies and how many credits each reply costs.

How do credits work? A Max reply is 0.25 credits, a Mini reply is 0.15, a Pro reply is 1 credit, an Economy reply is 0.5 credits.

See also: AI Model & BYOK · Billing Overview · Setting Up Your AI Bot