Input price

How it works

The input price is the rate charged for every token in the request, quoted on this site in USD per 1 million tokens. It covers the system prompt, user messages, earlier turns, attached files and images (after conversion to tokens), and tool definitions and results.

Most makers discount input tokens read from a prompt cache, and some charge a higher rate for very long prompts. Rates can also differ between the maker's own API and cloud platforms, and between standard, batch and priority tiers.

In chat and agent workloads the same context is re-sent on every turn, so input often outweighs output on the bill even though each output token costs more. Track input tokens per task, not just per call.

Worked example

At an illustrative $3 per 1M input tokens, a 20K-token prompt costs 20,000 / 1,000,000 x $3 = $0.06. An agent that re-sends that 20K context over 30 turns pays about $1.80 in input alone, before any caching.

Common mistakes

• 1Estimating cost per call rather than per task, missing how agent loops re-send context.
• 2Forgetting that tool definitions and images count as input tokens on every request.
• 3Comparing list input prices without accounting for cache discounts and long-context rates.