> ## Documentation Index
> Fetch the complete documentation index at: https://fireworks.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Usage and Cost

> Track estimated Claude Code session cost and model mix locally. Then view Fireworks model usage by model, user, or API key and Fireworks account cost by model.

| View | Shows | Open it |
| - | - | - |
| **Claude Code status line** | Live estimated session cost, model mix, cache share | Connect with `nexus claude` |
| **Session meter** | Requests and estimated cost for one Claude Code session | `nexus claude usage` |
| **Usage dashboard** | Fireworks model usage by model, user, or API key | [Account usage](https://nexus.fireworks.ai/usage) |
| **Spend Limits** | Current-period spend on supported Fireworks serverless models against each user's cap | [Spend Limits](https://fireworks.ai/docs/nexus/usage-limits.md) |

## Track one Claude Code session

`nexus claude` adds a status line with the estimated session cost and which models served it. An existing status line is not replaced.

![Claude Code status line with a session cost bar split across several models](https://mintcdn.com/fireworksai-docs/iVVGeuaohW3vcBl4/images/fireworks-nexus/session-cost/status-line.png?fit=max\&auto=format\&n=iVVGeuaohW3vcBl4\&q=85\&s=98c3361de6153b4876dbf302df109767)

Session cost, model mix, and cache share in Claude Code

The session meter only works for Claude Code, on the machine where the session ran.

* Other harnesses send requests through Nexus but don't have this local meter yet.
* Until the first billed call, the status line shows the model name. After that, it shows two lines:

| Line | What it shows |
| - | - |
| **1** | A spend bar and the estimated session cost. Each bar segment is a model that served traffic, sized by its share of estimated cost. On Nexus Auto (`auto`), once more than one model has served the session, it adds `routed model:` and the model that served the latest turn |
| **2** | Each model's name and cache share. When more than one model has served the session, each model's estimated cost too |

* Several models can appear in one bar. See [Nexus Auto](https://fireworks.ai/docs/nexus/auto.md) for why `auto` can use more than one model in a session.
* The auto-mode line below the bar belongs to Claude Code, not the Fireworks `auto` model ID.

## Inspect a session

Run `nexus claude usage`, choose a session, and open its live meter. It shows model choices, tokens, and estimated cost per request. Press `q` to quit.

| Command | Use |
| - | - |
| `nexus claude usage` | Choose a session, then open its live meter |
| `nexus claude usage --days 7` | Search further back than the default 3 days (1 to 365) |
| `nexus claude usage --session <id>` | Open one session directly. Takes a session ID or a log file path |
| `nexus claude usage --plain` | Print a one-time plain-text snapshot |
| `nexus claude usage --last-n 5 --json` | Return the five most recent sessions as JSON |
| `nexus claude usage --verbose` | Print one row per request, with rate details |

In the live meter:

| Key | Action |
| - | - |
| Tab | Focus the agents pane, when the session has subagents |
| ↑ and ↓, then Enter | Choose an agent in the pane and track it |
| Esc | Leave the agents pane, or return to the session list |
| `q` | Quit |

Each call is split into separate token buckets, each priced at that model's rate:

| Column | What it is | How it bills |
| - | - | - |
| `uncached` | Input tokens that missed the cache | Full input rate |
| `cached` | Input tokens served from the prompt cache | The model's cached-input rate. The discount varies by model |
| `write` | Tokens written into the cache | Anthropic charges a write premium. Fireworks models show 0 here |
| `cache%` | Share of input tokens served from the cache | Not billed separately |
| `out` | Generated tokens | Output rate |

## Break down Fireworks usage and cost

The [usage dashboard](https://nexus.fireworks.ai/usage) breaks down Fireworks traffic by model, user, or API key. The **Cost** view groups Fireworks account spend by model.

![Usage dashboard Cost view split across several models](https://mintcdn.com/fireworksai-docs/iVVGeuaohW3vcBl4/images/fireworks-nexus/session-cost/account-cost.png?fit=max\&auto=format\&n=iVVGeuaohW3vcBl4\&q=85\&s=119e099c2121467fd9fe75de5f98ca61)

Account cost by model. Names and amounts are illustrative.

The dashboard covers Fireworks model usage and cost only. It doesn't include usage billed by closed-model providers.

## Know which number you are reading

* **Status line and session meter:** a local estimate from token counts and published model rates.
* **Claude Code's built-in HUD:** Anthropic list prices.
* **Usage dashboard:** Fireworks model usage and cost for the selected range.
* **Billing page:** the source of truth for your Fireworks invoice.

- The session estimate prices Fireworks models at Fireworks rates, which `nexus model list` shows.
- Anthropic and OpenAI models are priced at provider list prices, from a cached copy that also works offline.
- Tool use, service tier, and region can make the invoice differ from the estimate.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.