After you tinker with MiocAI a bit (or a lot), you eventually come across this behemoth of a square with a severe case of purple acne. No worries, It's fine. In fact, it wants to be touched. Thats our model selector.

Model Identification

You can order a model in two ways. The Visual one, and the List one. Whats the difference? Not much. Just that its easier to tell which models are better using the visual selector.

Model colors

Models that are marked yellow with a coin in their corner are credit-charged models. Every model above 300 billion parameters falls into this band, and each generation draws from your balance according to two things at once, the size of the model and the number of tokens it actually produces.

The arithmetic is easy to hold in your head. Imagine a one trillion parameter model writing three hundred tokens. That reply costs exactly one credit. A two trillion parameter model producing those same three hundred tokens costs two, and a one trillion model that keeps going for six hundred tokens also costs two. The price therefore travels up or down along both axes, model muscle on one side and reply length on the other.

Since the true figure depends on the finished text, the picker cannot print one flat price for every message. It instead shows the rate for three hundred tokens, which anchors the entire curve. Multiply that rate by the number of three hundred token blocks the model writes, and you land on the final charge. The coin badge on the selected model mirrors this rate and shifts as you drag the circle between models.

Not to be confused with models that are yellow WITHOUT a coin, which means that they are currently unstable.

Rarely, models will be red, this inidcates excessive use, and therefore potential downtime on that model. Doesnt mean it neccesarily doesnt work, just means that a lot of requests to it failed recently.

Reasoning tokens

If you are on a plan that unlocks reasoning and you turn it on, the hidden blueprint the model works through before it answers is billed as well. Those reasoning tokens follow the rate of the model that produced them and they are folded into the same message, so one reply can carry a main charge and a reasoning charge together.

Apex never pays

Subscribers on the Apex tier skip all of this. Large models, reasoning, and image work stay included, and their coin displays fall to zero.

Overdraft and the credit gate

Credits are loose, not a brick wall in the middle of a sentence. When a reply costs more than you hold, it still finishes and your balance is allowed to slide below zero. What changes is the request after it. As soon as your balance sits at zero or lower, asking for a credit-charged model is declined before any work starts, and MiocAI offers the top up flow in its place. Clear the debt or add credits and the gate reopens. Balances are shown to two decimal places, so fractional costs remain visible instead of being rounded out of sight.

The free first message

A brand new conversation grants one complimentary generation on the auto selected flagship, even when that model would ordinarily bill credits. Once the chat has a couple of messages behind it, normal pricing returns.

Regenerations cost less

Rerolling an answer is cheaper than writing it the first time. Every regeneration takes forty percent off the credit cost of that message, and the discount covers the whole charge, the main reply together with any reasoning tokens attached to it. Where a normal response would have run one credit, the reroll lands at six tenths of a credit. The saving is applied automatically, so there is nothing to toggle and no code to redeem.

Model placement

Models placed further up usually are less capable than models further down. This does not neccesarily mean they are unusable, quite the contrary, they are completely usable. We get around 2 million generations on the lowest 4 models combined every day!

Models further down get progressively better at roleplay, creativity and reasoning however. With the very bottom littered with SoTA models.

We rank these models by: Unfiltered-ness, Size (intelligence), Creativity, and Reasoning.

Changes

With the volatile nature of LLMs at the moment, our model selection changes frequently. our most used models stay the same, but we often add/remove models. Its worth trying them!