The message this page prices
Swap in the sizes of your own sessions.
- Input tokens
- 6KCard, persona, lore and the history in the window.
- Output tokens
- 300Two or three paragraphs of reply.
- Input to output
- 20:1The ratio that drives everything below.
Where the money goes in one message
Priced on DeepSeek V4 Flash 0731.
| Item | USD |
|---|---|
Input — card, lore, history$0.0008 · 91% of chart maximum
Input is 91% of the message. Compare input rates first.
- Regenerating is cheaper than it feels.
- It re-sends the same prefix for a new reply, so it costs about what the original message did. Growing the prefix is the expensive action, not producing text.
A 1,000-message session
The same message, across the roleplay models at current catalog rates.
| Item | USD |
|---|---|
DeepSeek V4 Flash 0731$0.90 · 15% of chart maximum
Input tokens times the input rate plus output tokens times the output rate. Redo it with your own counts from the catalog or /pricing.
What actually lowers the number
In order of impact for this workload.
- Trim the prefix, not the reply.
- A lore entry that never comes up still costs you every turn. Cutting a thousand tokens from the card beats any output setting.
- Compare input rates first.
- Two models with the same headline price can differ severalfold on the side you are buying.
- Watch the window, not just the price.
- A thread that fits at message 20 may not at message 200, and it fails with a 400. Context windows are on every model page and on /models/long-context.
- Prompt caching is not in these numbers.
- A mostly static prefix is what cache reads make cheaper. Only rates the catalog publishes are shown here.
Keep going
- Roleplay modelsThe curated set behind these figures.
- Roleplay clientsSillyTavern, RisuAI and Agnai setup and privacy.
- PricingHow credits are priced.
- Cheapest modelsEverything under $0.35 per million input tokens.
- Capping unattended spendThe same arithmetic for a runaway agent.
- BillingWhat a usage record holds, and what happens at zero.