Costs and prices

RepoLantern is free. You pay only your AI provider, and you see the estimate before anything is spent.

Where the price comes from

For the model you choose, RepoLantern uses, in order:

1. Your price, if you entered one under Advanced.
2. A verified price, checked against the provider's price page, with its link and review date.
3. The community catalogue (LiteLLM's public price list, refreshed daily), labelled not verified. A price is only ever taken for the same provider and model.
4. Price unknown: the model still runs, but the spending limit can't apply; enter your price to use it.

The model step shows which one applies. RepoLantern never takes prices from web search or from resellers.

How cost is computed

For every call, from the token counts your provider reports: uncached input × input price, plus cache reads × cache-read price, plus cache writes × cache-write price, plus output (thinking included) × output price, per million tokens. These figures are checked live against each provider's own usage fields. Your provider's invoice is the authority: regional, contract and long-context prices can differ.

Before, during and after

• Before: the estimate shows likely, low and high cost; you approve it and set a spending limit.
• During: the live page shows the money spent against your limit.
• After: How it was made shows calls, tokens and cost by stage, and the price used.

Keeping it cheap

• Choose a small, fast model: documentation reads a lot and writes a moderate amount.
• File summaries are remembered by content: a second run on the same repository only pays for what changed.
• Beginner, senior and architect guides share one analysis.
• Prompt caching is used where it saves money (verified on OpenAI and Claude).
• Thinking is set per step: little while reading, more only where writing needs it.