A month of small coding sessions, costed both ways

Money mattersEN2026-10-27

A month of coding help rarely looks like a month of continuous work. There are short questions, a few stubborn bugs, and days when the assistant stays closed. That uneven pattern is worth measuring before choosing between a monthly subscription and an API bill.

This notebook uses a worked example: 18 active days, six small sessions per day, and no usage on the remaining days. It is not a report from a customer account. The activity and prices below are explicit assumptions, so you can replace them with your own numbers.

§ Write down the work first

Our example month contains 108 sessions. A session might involve explaining a function, suggesting a patch, or checking a test failure. For the arithmetic, each session uses an average of 20,000 input tokens and 2,000 output tokens.

Tokens are pieces of text. Input tokens include your instructions and any code, conversation history, or tool results sent to the model. Output tokens cover the response it generates. A short question can therefore have a large input bill if the coding tool attaches a long conversation or several files.

Active days: 18.
Total sessions: 18 × 6 = 108.
Monthly input: 108 × 20,000 = 2,160,000 tokens.
Monthly output: 108 × 2,000 = 216,000 tokens.

These averages simplify the calculation. In practice, record the usage reported by your tool or provider rather than estimating it from the length of your prompts.

§ Price the month as a subscription

Assume a hypothetical subscription costs $20 per month. The month costs $20 whether you use it on all 18 active days or open it only once. Across 108 sessions, that works out to about 18.5 cents per session.

That does not make the subscription wasteful. A fixed payment can be convenient, and a subscription may include an interface or features you value. But check its usage limits and whether it includes API access. A chat subscription should not be assumed to pay for requests made through a separate coding tool.

§ Price the same month by tokens

Now assume an API rate of $2 per million input tokens and $10 per million output tokens. These are illustrative rates, not a quote for Vana API or any particular model. Use the current rates for your selected model when making your own comparison.

Input cost: 2.16 million × $2 = $4.32.
Output cost: 0.216 million × $10 = $2.16.
Total token cost: $6.48, or six cents per session.

Under these assumptions, token billing costs $13.52 less than the subscription. At the same average session size and rates, about 333 sessions would cost $20. That threshold changes as soon as the model, context size, or response length changes.

§ Keep the comparison honest

The simple calculation treats all input tokens alike. Actual billing may distinguish cached input, cache creation, or other token categories. Coding agents can also make several requests during what feels like one session. Use the final usage records and applicable rates, not just the visible answer.

Compare equivalent work where possible. A month using Claude Haiku 4.5 is not automatically equivalent to a subscription offering different models, limits, or tools. Cost matters alongside whether the assistant completes the task reliably.

§ Try a small measured month

Vana API provides one endpoint at https://api.vanaapi.com, with OpenAI compatibility plus Anthropic Messages and Gemini native protocols. It works with Claude Code, Codex, Gemini CLI, and OpenAI SDKs. Billing is per real token, with no monthly fee.

If you want to measure your own pattern, create a Vana API account and start with a small, ordinary workload. Keep a note of active days, token totals, and the final charge. After a month, compare that bill with the subscription you would actually buy. The useful result is not a universal winner, but a cost you can explain from the work you did.