How Are Assistant Instructions and Daily Charges Billed in the Assistants API?

Below is a summary of how the pricing works when you integrate Assistants via the API:

Instructions and Token Billing

• Instructions as Context:
  When you create an assistant with specified instructions (often provided as a system message or similar), those instructions typically become part of the context provided to the model on every call. That means, if you have a long-running conversation where the assistant’s instructions are preserved as context, they are included in the message history.

• Per-Call Token Usage:
  Every call you make to the API is billed according to the number of input tokens and output tokens that are processed. So if your conversation history includes the original instructions along with messages A and B, and then you send a new message C, the input tokens will include the instructions along with messages A and C (and any other context being carried over). The response (such as message D) will be billed based on the output tokens returned.

• Extra Charges:
  There are no separate charges specifically labeled as “instructions charges.” The cost simply comes from the total token count, meaning the cost of the instructions is already baked into the per-token pricing.

Additional Daily Costs

• gpt-4 Daily Charge:
  For Assistants using gpt-4, there is an additional charge specified as $0.20 per GB per day. This cost is generally associated with persistent usage (for instance, caching or state management) and is applied alongside the per-token costs.

• Combined Billing Example:
  Using your scenario as an example:

  • When you send message A and receive response B, you’re billed for the tokens contained in those messages (including any persisted instructions if they’re sent as part of the input).
  • When you later send message C (with the conversation history including instructions and message A) and later receive response D, the token usage is summed from all included messages.
  • The daily cost would be the sum of:
    • The token-based charges (from inputs and outputs), plus
    • The fixed $0.20/GB per day cost for using gpt-4.

In Short

• Yes, the instructions are sent along with your conversation context each time if you have configured your assistant that way, and you are billed for the tokens that the instructions consume.
• There are no extra charges beyond those token counts, aside from the separate daily charge (e.g., $0.20/GB per day) which applies to the persistent use of gpt-4.

This means your daily billing would indeed include both the per-token usage (covering messages A, B, C, D, and the instructions) and the daily operational fee.