You never pick a model. Zinley reads every task and puts the right amount of intelligence on it — so light work barely touches your credits, and hard work gets a team.
Every AI product makes you choose: a frontier model that burns your quota on small talk, or a cheap one that fumbles the work you actually care about. Zinley refuses the choice. Auto reads each task and matches it to the least intelligence that will do it well — and when a task deserves more, it brings a team: a boss that plans and judges, and workers that grind in parallel. The goal is simple. Make your credits go as far as they can, and never put second-rate intelligence on first-rate work.
“Thanks, sounds good!” does not need a frontier model. A contract review, a twenty-source research job, or a refactor across fifteen files absolutely does. Bill both against the same brain and you either overpay for chit-chat or under-serve the real work.
Zinley bills credits at the real cost of the work — so the routing is your quota. Every message that lands on a lighter model is credits you keep. Every hard task that lands on the team is quality you didn't have to pay a flat premium for.
Light work runs light so heavy work can run heavy. More total usage from the same credits, and better answers on the tasks that deserve them. That's the whole trade.
Before anything runs, Auto weighs the task — in a fraction of a second, before you can notice. A quick reply gets a light touch. Real work brings the team. Scheduled jobs run quietly on efficient models, so your morning brief never eats your afternoon's budget.
Figure 1. Every task is weighed before it runs. Small things stay small, real work gets the team, and background jobs never make noise in your balance.
You'll never see the seams. A conversation doesn't jump between brains mid-thought, and stepping up is instant the moment the work turns serious. And when Auto isn't sure how hard a task is, it rounds up — doubt resolves toward quality, never toward savings.
Real work runs like a real company. A frontier boss model reads the brief and does the thinking that matters — the plan, the judgment calls, the final word. The legwork goes to workers that run in parallel: researching, reading, drafting, building.
The team fans out, the results come back, and the boss signs off on one answer. On Pro and Max, when a piece of the job turns out genuinely hard, it escalates to a premium worker — quietly, only when it matters.
Figure 2. The boss thinks; the team grinds in parallel. Everything converges into a single answer the boss stands behind.
The expensive brain never does the busywork. It thinks, delegates, and reviews — while the long grind runs on models that cost a fraction as much. Frontier judgment on every task, without frontier rates on every token.
Every plan runs Auto. What changes with your tier is who shows up when the work gets hard:
The “Elite” brain is the strongest model we run, and it's part of Pro and up — you can't unlock it with credits. Seat it as your boss with one switch in settings: the team stays the same, and the one steering it gets sharper.
You're billed for the real cost of work, so every routing decision shows up in your balance. Four things quietly work in your favor:
Quick replies, titles, summaries, housekeeping — all handled by fast, efficient models. The frontier brain never bills you for small talk.
In a big job, most of the tokens are legwork — and the legwork runs on workers that cost a fraction of the boss. Most tokens in a hard task are cheap tokens.
Zinley reuses what it has already read instead of paying full price to read it again. The deeper you go in one thread, the less each turn costs.
No model picker, no dials to watch. The routing works in your favor by default, on every task, whether you think about it or not.
Quality is one axis. Speed is another. On Pro and Max, you can switch on Priority processing: supported requests route through a faster priority lane, so replies land quicker when you're on the clock.
It costs a premium rate — and here's the part we're proud of: you're only charged the premium when priority is actually delivered. If a request ends up served through the standard lane anyway, you pay the standard price. No paying for a fast lane you didn't get.
Standard lane, standard rate. The default — and all most people ever need.
Faster lane on supported requests, premium rate — billed only when the fast lane actually serves you.
It's a switch, not a plan change. Flip it on for the deadline sprint, flip it off after lunch. Off by default, change it any time in settings.
You just talk to Zinley. Behind every message, Auto weighs the task, a boss decides what deserves its own attention, and a team does the legwork in parallel. Light things stay cheap so your credits last; hard things get the best available brain so the answer holds up.
And before big work reaches you, it's torn apart by independent reviewers — but that's a whole other page.