How are lines of code converted to tokens?

Code is denser than prose. The estimator uses roughly 8-12 tokens per line of code depending on language (verbose languages like Java are higher, terse ones like Python lower). This is an approximation — exact counts vary with formatting and comments.

Why does review cost less than generation?

Review and refactor read the full codebase as input but generate far fewer output tokens (comments, diffs, suggestions). Generation produces large output, and output tokens are typically 3-5x more expensive than input.

Does this account for multiple passes or context windows?

The estimate assumes a single conceptual pass. Real agentic coding tools re-send context many times, so multiply by your expected number of iterations for a realistic agent budget.

Is my code uploaded anywhere?

No. You only enter a line count and settings. Nothing is sent to any server — the calculation runs entirely in your browser.

Code Generation Cost Estimator

Estimate the LLM cost of a coding task

Before you point an AI model at a repository to generate, review, or refactor it, you want a rough idea of the bill. This estimator turns a line count and a language into token estimates, then prices them against GPT-4o, Claude, and Gemini so you can budget a coding task in seconds.

How the estimate works

Source code is token-dense — far denser than prose. A line of code averages roughly 8-12 tokens depending on the language: verbose languages like Java or C# sit at the high end, terse ones like Python or Ruby at the low end. The estimator multiplies your line count by a per-language density factor to get input tokens, then applies a task-specific output ratio:

input_tokens  = lines × tokens_per_line
output_tokens = input_tokens × task_output_ratio
cost = (input/1e6 × in_price) + (output/1e6 × out_price)

Generation produces a large amount of output (you’re writing new code), so its output ratio is high. Review reads everything but emits only findings, so its output ratio is small.

Tips for a realistic budget

Agentic coding tools rarely make a single call — they re-read files, retry, and expand context, so real spend is often 3-10x a single-pass estimate. Multiply accordingly. To cut cost, prefer cheaper models (GPT-4o mini, Claude Haiku, Gemini Flash) for review and bulk transforms, and reserve frontier models for the parts that genuinely need reasoning.