---
title: Cost tab
description: See the tokens and estimated cost of the language model calls in one run.
editUrl: true
head: []
template: doc
sidebar:
  hidden: false
  attrs: {}
pagefind: true
draft: false
---

What a run's language model calls used and roughly what they cost, split by provider and model. Use it to see which model drove a run's spend, or to check that cached input is being picked up.

![The Cost tab of a run that called one model, with markers on the Provider / Model column (1) and the Cost column (2).](../../../../assets/screenshots/web-app/instance-cost-tab.png)

1. Provider / Model - the model each row covers, with its context window.
2. Cost - the estimated cost in US dollars.

The tab only appears when the run recorded token usage on language model spans from an integration Prefactor reads tokens from: the OpenClaw plugin, the AI SDK, and GitHub Copilot Chat. The same totals appear as the **Cost breakdown** cards above the tabs on the [Agent instance page](/web-app/agent-instance#cost-breakdown).

The table has one row per provider and model, then a **Total** row. **Provider / Model** uses the names from [models.dev](https://models.dev) when it knows the model, with the model's context window beside it, such as "200k context". **Input**, **Input (cached)**, and **Output** are token counts. The AI SDK doesn't report cached input, so its rows show 0 there.

**Cost** is an estimate in US dollars, worked out from models.dev's published prices for each model. Every non-zero cost starts with `~` to show it's approximate. "free" means the model's price is zero, and "—" means models.dev has no price for the model. When one row shows "—" and another has a cost above zero, the **Total** cost shows "—" too, because a partial sum would understate the spend.

## Related

- [Agent instance page](/web-app/agent-instance) — the cost breakdown cards and the strip above the tabs.
- [Agent instance page › Activity tab](/web-app/agent-instance/activity) — token counts on each LLM call card.
- [Span](/platform/concepts/span) — what a span records, including the LLM spans the tokens come from.