People sometimes ask why AverTales sells credits instead of a flat monthly plan with unlimited messages. It is a fair question, and the answer is not that we prefer it. AverTales is built for serious roleplay: large casts, long campaigns, a world that remembers, and the experience put first. An unlimited plan can only be paid for by quietly making that story smaller. This page shows the arithmetic so you can judge for yourself, and where the monthly plan we do offer fits into it.
AverTales is built for serious roleplaying: big casts, campaigns designed to run for more than a thousand turns, and a world that keeps track of all of it. Every story turn sends that world to a frontier AI model, and the model's owner charges us for every word it reads. A long story with a capable model costs real money on every single turn. A flat fee can only cover that by capping how much the model is allowed to remember, routing you to a smaller model, or hoping most people play very little. We would rather charge you exactly what a turn costs, show you the price before you send it, and never cut the story behind your back.
Credits are that exact charge, made visible. You are not paying for a promise of "unlimited". You are paying for the actual turns you play, on the model you chose, at the size your story has grown to, with nothing hidden. The experience comes first; the price simply follows it.
There is also a monthly plan, and it is not an exception to any of this. It includes a set number of story turns a day on our compact storytellers, which cost little enough per turn to fit inside a fixed price at full context, and it adds credits every month for everything else. Nothing about the story is made smaller to pay for it. How it fits is further down.
1,000 credits = $1. A credit is a fixed slice of a dollar, which is what makes everything else on this page work: the price of a turn can change with the model market, but the value of a credit never does. The Starter pack is sold at face value; the larger Plus and Mega packs add bonus credits on top, so the bigger the pack, the more each dollar buys.
A story turn is priced from three things: the model you picked, how much story the model has to read for that turn, and our margin over the raw model cost. What the model reads is not just your last message. It is the recent turns, plus the memories and lore cards recalled for that moment: the characters in the scene, the relationships, the promises made, the places revisited. Longer stories and bigger casts carry more of that, so the same turn costs more at turn 400 than at turn 4, and premium models cost more per word than compact ones. You always see the estimate next to the send button, and the exact charge on each turn afterwards.
The reply length is yours to choose, and the ceiling is generous. Pick a target from a short conversational beat up to a long scene, or let the narrator choose the natural length for the moment, well past a thousand words when the scene calls for it. That matters with a big cast: many platforms cut every reply at a few hundred words, and a reply that stops mid-scene means some of the characters in the room never get a line. Here a crowded scene gets the room it needs, and the only cost is a few credits more on the length surcharge.
| Thing | How it is charged | Why |
|---|---|---|
| Story turn | Metered by model and how much story is read: recent turns, recalled memories, lore cards | This is the real cost. It is the one thing we cannot flatten without shrinking the story. |
| Story turn on a plan storyteller, for subscribers | Free, up to the plan's daily number of turns. Regenerations and OOC questions count as turns. | A compact storyteller costs little enough per turn to include a daily number in a fixed price, with the full story still read. |
| What the model writes back | Small flat surcharge by response length, which you choose | Output is a small share of a long story's cost, so a few credits by length preset covers it without a second meter. The length ceiling is generous, not a tight cap. |
| Memory, summaries, the Director | Inside the turn price | The side calls that keep the world consistent are not charged separately. You never pay extra for remembering. |
| Secret info (narrator notes) | Small flat surcharge per turn while enabled | The Director does extra work to keep hidden knowledge and future events straight, so it carries a small fixed fee. |
| OOC chat | Metered at cost, no margin | Asking about your story should not cost like playing it. |
| Illustrations, backgrounds, music, video | Flat price per item, never charged for a failure | These are optional extras with a predictable cost, so they get a predictable price. |
| Daily free credits | Free, every day | Anyone can play a little without paying. There is no trial that converts to a charge. |
Current per-item prices, packs, the monthly plan and the daily free amount are on the credits page. They are settings we tune, so this page describes the shape rather than quoting numbers that may drift.
Here is the arithmetic that rules it out. Take one dedicated player doing 50 turns a day, which is a normal evening of roleplay rather than an extreme one, and two sizes of world: a moderate one where each turn reads about 80,000 tokens of story, memories and lore, and a large one at about 150,000.
Most of what a turn reads is the same as the turn before, so providers serve that part from a prompt cache at a fraction of the input price, and AverTales lays out its prompts so as much as possible can be cached. We assume 60% of each turn is read from cache. That is deliberately below the 85 to 95% we measure on a warm turn, because a turn after a break starts cold, every few turns the recent-story window moves forward and part of the prompt is read fresh, and not every provider's cache holds. These are raw provider costs before we add anything, and they count only the narrator reading the story: not its reply, and not the side calls.
| Model class | Provider input rate | Cache read rate | Moderate world, 80K context | Large world, 150K context |
|---|---|---|---|---|
| Compact model | $0.14 per million tokens | $0.003 per million | about $7 | about $13 |
| Full-size model (DeepSeek V4 Pro) | $1.20 per million tokens | $0.12 per million | about $66 | about $124 |
| Premium model (rarely used here) | $3.00 per million tokens | $0.30 per million | about $166 | about $311 |
Note that even the moderate world is already bigger than most competitors allow at all: the flat-fee platforms below cap context at a few thousand to 32,000 tokens, so 80,000 is more than double their top tier, and 150,000 is out of reach.
Caching changes the compact row the most. A compact model is now cheap enough to sit under a flat fee, which is exactly why flat-fee platforms run small models. The full-size model is where it breaks. A dedicated player on DeepSeek V4 Pro costs about $66 to $124 a month in narration input alone, several times a $20 plan, before the reply, the Director, the memory pass or payment fees. Even at the best cache rate we have measured, around 90%, that player still costs about $27 to $51, more than a $20 plan brings in. The premium tier, which almost nobody here needs, is worse still. A flat fee only survives if the people who play a lot are subsidised by the people who barely play, and the business depends on the second group staying large. That is not a plan, it is a prayer.
The sites that advertise unlimited messages make the numbers work in one of three ways, and usually all three: they cap the context so the model can only see the last few thousand words of your story, they run small self-hosted models instead of frontier ones, and they throttle quietly when someone plays too much. Each of those is a cut to exactly the things people praise AverTales for: the long memory, the model choice, the sense that the world remembers.
Unlimited turns are pointless if every turn is a worse turn. We could offer an unlimited plan tomorrow by capping context at 32K, dropping the premium models and shrinking the memory pipeline. It would look generous on a pricing page and play like a lesser game. We would rather sell you the real thing at an honest price.
The same table shows where a fixed price can work. A compact storyteller costs a small fraction of what a full-size one costs per turn, so a set number of turns a day on those models fits inside a monthly price without touching what the model reads. That is the whole of the monthly plan:
What the plan does not do is promise "unlimited". Its limit is a number you can see before you subscribe, on storytellers we name, and past it a turn costs what it costs anyone. That is the difference between a plan with a visible limit and an unlimited plan with a hidden one.
The plan is optional. Credits work exactly as before without it, and you can cancel at any time and keep the plan to the end of the month you paid for.
The plan's price, daily turns, monthly credits and included storytellers are on the credits page. Like every other price here, they are settings we tune.
The second reason is that the cost underneath us is not stable. Model providers change their prices, sometimes sharply and with a few days' notice. One recent example: DeepSeek V4 Pro, a model many people built on.
| Period | Input | Output | Change |
|---|---|---|---|
| Launch, late April | $1.74 | $3.48 | |
| Late May to mid August | $0.435 | $0.87 | 75% promotion made permanent |
| From 16 August, off-peak hours | $0.66 | $1.98 | Promotion ends, peak pricing begins |
| From 16 August, peak hours | $1.32 | $3.96 | Roughly 3× the summer price |
Through the gateway we use, the input rate for that model went from about $0.44 to $1.10 per million tokens overnight, a 2.5× jump. Anyone who had promised unlimited turns on it in July was paying two and a half times more for the same promise in August, with no way to raise their price mid-month and every incentive to start cutting quietly.
It also runs the other way. Newer models arrive that are cheaper and as good, older ones get dropped by hosting providers, and the few providers left standing for a popular old model can charge what they like. The ground moves every month.
The turn simply costs more credits on that model, or fewer, from the day the provider changes it. Your credits keep their value. You can switch to a different model in one click, and you see the new estimate before you send. Nothing is cut, nobody is throttled, and no promise gets broken, because the promise was never "unlimited". It was "you get exactly what you paid for, at full quality".
None of this is a criticism of the other platforms. Each has made a sensible choice for what it is. The point is that a flat fee always comes with a trade, and it helps to see where the trade sits.
The subscription tiers are, quite literally, context window sizes: from a few thousand tokens on the free plan up to 32K on the top plan, and on some models you spend credits every turn to keep the context up. That is what an unlimited plan looks like once the bill has to fit a fixed price: the model is not allowed to remember much of your story.
Unlimited turns, short memoryThese run their own or small hosted models with capped context, which is the only way a flat subscription stays solvent. It works well for quick chats and companions. It is a different product from a long campaign with a cast of characters and a world that remembers.
Unlimited turns, smaller modelsNomi charges around $16 a month for unlimited messages, and makes that work by running a single model it built and trained itself for companionship, with a cap on companions and on daily images. That is a smart way to make a flat fee solvent: a model you own costs the same every month, whatever the market does. The choice on offer is between its current generation and the previous one, both Nomi's own; there are no frontier models from other labs to switch to.
Unlimited turns, one modelThe one platform built the way AverTales is built, with real third-party models and generated visuals, arrived at credits as well, with no subscription at all, and a developer who says outright that a flat plan would need "either predatory pricing or external funding". The difference is visibility: they do not name the model behind their tiers, and their own wiki says the cost per turn is impossible to predict without testing. Here you see the model, the estimate before you send, and the exact charge after.
Same model, less transparentIf you are a power user, this is the cheapest way to do any of this. It is free, you bring your own API key, you pick your own provider, and you would likely spend about half of what you spend here, because nobody is in the middle. The price is your time: you host it, fund it, tune it, and build your own memory out of extensions you maintain by hand. It is a toolbox rather than a game. The full comparison lays it out side by side.
Cheapest, most workSo the honest map is this. Want it as cheap as possible and don't mind playing sysadmin? SillyTavern is a great choice. Want unlimited turns on a flat fee? AI Dungeon and the others do that well, and the trade-off is a smaller model and a shorter memory whether or not it is written on the box. Want the long-context, full-quality version and would rather just play than set things up? That is what AverTales is for. Credits are how we can offer it without quietly trimming it to make the numbers work, and the monthly plan covers everyday play on the compact storytellers without changing that.
A credit system can be abused too, so here is what we hold ourselves to.
Almost always because the story grew. Every turn sends the model your recent story plus the memories and lore it needs, so a turn at chapter twelve reads more than a turn at chapter one. Switching to a pricier model, choosing a longer response length or turning on secret info also raises it. The estimate next to the send button tells you before you commit.
You have four dials, and none of them shortens the story itself.
Switch models for the turns that matter less. Use a compact model for travel, shopping and small talk and save the full-size one for the scenes you care about. The model picker shows each model's relative cost, and you can change it between any two turns.
Set your lore and memory limits. Your settings control how many lore cards and memories are recalled per turn. A lower standing limit makes every turn cheaper; a higher one gives the model more of the world to work with when you want it.
Trim or boost a single turn. Type /trim-N in your action to recall N fewer cards for that turn only, or /boost-N to recall N more when a scene needs the full cast. The composer has a chip for the same thing. Your standing limits are untouched.
Turn off the extras you are not using. Fewer auto-illustrations per turn, a shorter response length and no secret info all take a few credits off each turn.
Story turns are charged at a little over twice the raw provider cost of the input. That margin is what pays for the memory pipeline, the Director, summaries, hosting and payment fees. Because the model's output carries only a small flat surcharge, the side calls are not charged at all and OOC chat is sold at cost, the effective margin on a whole session is lower than that headline number. Pure API resellers charge less, and they sell you nothing but the tokens.
Purchased credits never expire. The daily free credits are a top-up to a floor rather than a stack, so they do not accumulate, but they also never take anything away from what you bought.
A story turn is charged only after the narration succeeds, so a failed turn costs nothing. Backgrounds and music are the same: charged only once a track or backdrop actually arrives. Illustrations and video are charged up front, because the provider bills us either way, and refunded automatically the moment a failure is reported.
The credits from that pack are removed again, including any purchase bonus, in proportion to the amount refunded. If you had already spent them your balance can go below zero until the next purchase. That is the honest mirror of the refund and it appears in your credit history like everything else.
Yes. It gives you a set number of free story turns every day on the storytellers included in the plan, plus credits with every monthly payment for everything else. It is optional: credits work the same with or without it. The price, the daily number and the included storytellers are on the credits page.
Because of the arithmetic above. A compact storyteller costs a small fraction of a full-size one per turn, so we can include a daily number of turns on it in a fixed price and still send the whole story. A full-size or premium storyteller costs several times more per turn, and no fixed price covers that without capping something. Those stay available to subscribers, paid with credits.
Nothing stops. The next turn shows its price in credits before you send it, the same as for someone without a plan, and the plan's monthly credits can pay for it. The free turns come back at the daily reset.
Because there is no price that covers an unbounded cost. A $50 plan loses money on a premium-model player just as surely as a $20 one, only later. The only unlimited plans that work are the ones where the limit has been moved somewhere you cannot see it.