LLM API Token Cost Calculator

How do API usage fields map to the cost calculator?

Check the response protocol, then enter input and output tokens separately. Do not put total_tokens in both fields or estimate billed usage from visible answer length. This mapping does not establish that every Link model returns complete usage.

Output totals may include hidden reasoning or formatting tokens; do not add included details again. If usage is missing, check usage logs instead of entering zero for unknown values. Choose Per-task usage to project from one task, or Aggregated usage for totals across requests without multiplying by task count again.

Combine only billing categories for the same model and effective rates. Retries already included in the totals are not added again; this does not mean no retries occurred. Aggregated mode does not infer actual request counts or per-task costs. Calculate different rates and cache categories separately, then sum them.

How do you convert per-thousand to per-million token rates?

All price fields use rates per million tokens. Multiply a per-thousand rate by 1000: for example, 0.0025 per thousand equals 2.5 per million in the same currency. This is a unit-conversion example, not a quote. Use one currency for input and output, and do not reapply a group multiplier already included in the effective rate.

Can cache reads and writes use the ordinary input price?

Do not assume so. This tool does not automatically separate cache rates. Check whether the protocol includes cache reads and writes in its input total, then price non-overlapping ordinary input, cache reads, cache writes and output separately. Split writes by retention period when necessary. Sum the categories and confirm unknown fields or rates with your provider first.

For example, if 3000 total input tokens include 2000 cache reads and no cache writes, ordinary input is 1000. Do not charge all 3000 at the ordinary rate and then add 2000 cached tokens. Use this split only after confirming the field relationship; do not apply OpenAI field semantics to Anthropic or another gateway without checking.

Field references: OpenAI Chat Completions usage · OpenAI token-counting guide · OpenAI cache-usage guide. Official protocol documentation does not guarantee Link pricing or compatibility.

Review the reproducible offline cost formula and input fields

GPT and Claude API cost example: 1,000 tasks

These are illustrative values, not AIFast Link or model-vendor prices: $2.50 per million input tokens, $10 per million output tokens, 3,000 input and 800 output tokens per request, and 1,000 tasks. Assume 10% additional billed retries with the same token usage.

(3000 × 2.5 + 800 × 10) / 1000000 = $0.0155
$0.0155 × 1000 = $15.50
$15.50 × 10% = $1.55
$15.50 + $1.55 = $17.05

The base cost is $15.50, retries add $1.55, and the total is $17.05. Exclude failed requests that were not billed. Calculate retries separately when their token usage differs. Separate cache, tool and image charges are outside this formula.

To reproduce the calculation locally, see Link GitHub examples and tests. The offline version uses Node.js 22 or newer, has no third-party dependencies and needs no network or API key. Included inputs are illustrative, not current offers.

Monthly plans reset daily, so the advertised monthly total cannot replace a pay-as-you-go estimate. Compare actual usage days with the Link pay-as-you-go and monthly plan rules.