Billing Questions

Why do I see multiple billing records after sending only one message?

A single operation in an agent tool may trigger multiple real model calls behind the scenes.

For example:

• The tool may first ask the model to understand the task.
• It may call another model request to read files or plan edits.
• It may call the model again after tool execution.
• Some clients also retry or split long work into several requests.

Why did I choose gpt-5.5 but see gpt-5.4-mini in billing?

Some clients, agents, or routing policies may call a smaller model for auxiliary work, such as classification, planning, summarization, or short verification.

This does not mean your main model selection failed. It means part of the workflow used another model for a supporting call.

What are cache reads and cache writes? Why are their prices different?

Many large-model providers use prompt caching.

• Cache write: the provider stores reusable context, such as long system prompts, project files, or previous conversation context.
• Cache read: the provider reuses already cached context.

Cache writes usually cost more than cache reads, because the provider needs to process and store the context. Cache reads are usually cheaper.

Why do some requests have few input tokens but still cost more?

Possible reasons include:

• The request triggered cache writes.
• The model used has a higher unit price.
• The client made multiple hidden calls around the visible action.
• Output tokens or reasoning tokens were higher than expected.

Check the model ID, input tokens, output tokens, cache read tokens, and cache write tokens together before judging the cost.

Why are expected charge and actual deduction different?

Expected charge is calculated from the model usage details. Actual deduction can differ because of rounding, discounts, cached billing rules, account-level settlement rules, or provider-side adjustments.

If the difference looks abnormal, keep the request ID and contact support.

Why does my balance seem to be consumed faster over time? How can I reduce usage?

Agent tools often send larger context as the project grows. Long conversations, more files, and repeated tool calls all increase token usage.

To reduce cost:

• Start a new conversation when old context is no longer needed.
• Avoid asking the agent to read the whole project unless necessary.
• Prefer smaller models for simple tasks.
• Use streaming for better perceived latency, but remember that streaming itself does not reduce token usage.
• Watch cache write records. Frequent cache rewrites may increase cost.

When should I contact support?

Contact support if:

• You see requests you did not make.
• A model ID is clearly not the one you configured.
• Actual deduction is far from the usage details.
• Requests fail but still appear to be charged.
• You need help understanding cache or token records.

Please include the request time, model ID, request ID if available, and a screenshot of the billing record.