What a token is

Models do not read letter by letter. They read tokens: a common word like "chat" is often one token, and a longer word can be two or three. A rough rule for English is 1,000 tokens is about 750 words. Other languages usually take more tokens for the same sentence.

The model company charges for text you send in and text it sends back. The reply is often the expensive part, so a short answer costs less than a long one. On ChatNexus that cost shows up as credits, not as a separate OpenAI or Groq invoice.

How ChatNexus charges

Each plan includes a number of bot replies (messages) and a pool of credits. Simpler models use fewer credits per reply. Larger ones use more. You stop when either the messages or the credits run out, whichever comes first.

On the Free plan that looks like this:

• 2 agents
• 300 messages a month
• 300 credits a month
• 30 days of access from the day you sign up
• Standard models only

A standard model often costs about 3 credits a reply. 300 credits is about 100 replies, so the credit pool runs out before the 300-message cap. A much larger model can cost around 25 credits a reply. The same 300 credits would only cover about 12 replies.

Starter can use models up to GPT-5.4 Mini. GPT-5.4, GPT-5.5, and GPT-5.6 need Pro or Enterprise. On those paid plans you can allow extra AI usage after the included credits, up to a monthly spending limit you set in Settings. Free accounts do not have that option. When the limit is hit, the bot stops.

There is no automatic "use a cheap model for easy questions" switch. You pick the model on the agent.

What gets sent on every reply

Each answer is built from a few pieces of text:

• Your system instruction, sent every time
• A few matching chunks from the knowledge base
• The last 4 messages in that chat
• The new question
• The reply itself

ChatNexus already limits history to those last 4 messages, so an old long chat does not keep growing the bill forever. Knowledge base files are split into chunks of about 1,500 characters. A question pulls a few of those chunks, not the whole file.

You cannot set a token cap or a custom chunk size in the dashboard. These are the levers you do have:

1. Shorten the system instruction

It is sent on every message. This is too long:

You are a helpful, friendly, and professional customer support assistant for
Acme. Your job is to help users with their questions about our products and
services. Always be polite and professional. If you don't know the answer,
tell the user you don't know. Never make up information that isn't in your
knowledge base.

This says the same thing:

You are Acme support. Answer only from the provided context.
If you are unsure, say so. Keep it short.

2. Write the knowledge base in short sections

A heading plus one clear answer works better than a 2,000-word brochure. The bot can then pull the returns paragraph without also pulling the whole about page. See how to write a knowledge base.

3. Tell it to keep answers short

Add a line like "Answer in a few sentences unless the visitor asks for steps." Long replies use more credits because the model is writing more text.

4. Pick a smaller model when you can

A FAQ bot usually does not need the largest model. Free accounts are limited to standard models. Move up only when the answers are actually wrong, not just because a bigger name sounds safer. A simple FAQ bot, like one built from the customer support FAQ template, is a good place to start on a standard model.

Voice replies cost extra too. See how ChatNexus caches repeat voice answers to keep that spend down.

5. Check the dashboard

Messages and credits used this month are on the dashboard and in Settings. Pro and Enterprise also show token totals, split into text sent in and text sent back, plus a 6-month trend. That is account-level usage, not a per-chat token report.