SpicyChat models, memory and tokens
SpicyChat can keep an old conversation on your account without including every message in every new reply. What the model can see right now is the context window. The plan and the selected model both affect its size.
What tokens mean in SpicyChat
Tokens measure pieces of text. They aren't a count of messages or a balance you spend for each reply. The context includes recent turns plus items such as the character definition, persona and relevant memories. A long character description leaves less space for previous messages on the same plan.
The documented reply ceilings are 180 tokens for Free and Get A Taste and 300 for True Supporter and I'm All In. A reply can end well before its maximum. If a character repeats itself, raising that ceiling won't necessarily fix the cause.
Why 32K is not on every model
Free and Get A Taste list 4K context, and True Supporter lists 8K. I'm All In supports up to 32K on compatible models. The smaller limit wins: a 16K model runs within 16K even when the plan allows more. For example, the documented SpicyXL option has a 16K model limit; it isn't the same as selecting one of the 32K models.
Change the model and compare one reply
In a hosted chat, open the conversation menu, then Generation Settings and Change Model. The chooser shows what your account can use. If a model such as Tamesis no longer appears, check that list for your account and the provider's release notes before assuming it is an outage. Models can move or be retired. If you are comparing two options, give each the same scene and one follow-up question. Changing the character, persona and prompt at the same time makes it hard to tell what the model changed.
Help the scene keep track
If a conversation forgets which person has a key, restate that detail in one line. Use the Memory Manager on eligible plans to pin a durable fact and Lorebook entries for world details that belong only to some scenes. See pricing and limits before buying more context just to fix one ambiguous setup.