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

# Model prices

> What the wallet pays: each model per million tokens (input, output, cached input), and each tool call

These are the list prices: per model for its tokens, and per tool call for the
work your agent does on its machine and in connected apps. They are what [extra usage](https://docs.splox.io/account/extra-usage.md)
charges the wallet once a plan window is full, and what a call costs when the
wallet pays for it. Inside your plan's windows you are not charged per token, and tool calls are free;
the windows are what the plan buys (see [Usage](https://docs.splox.io/account/usage.md)).

Token prices are in US dollars per **one million tokens**.

| Model | Input | Output | Cached input |
| - | - | - | - |
| `theo-quazar-1.3` | \$9.00 | \$18.00 | \$0.20 |
| `theo-spark-1.3` | \$4.00 | \$12.00 | \$0.20 |
| `kimi-k3` | \$3.50 | \$10.00 | \$0.30 |

## Tool calls

A tool call is the agent using its machine: a command run in the terminal, a file
read or written, a page opened in its browser, a call into a connected app. The
per-call price is what the machine's work costs, the way tokens are what the
model's work costs.

**On a subscription, tool calls are free.** Inside your plan they are never
charged, however many the agent makes.

Without a subscription, or when the wallet pays because a plan window is full,
each tool call that succeeds is charged **\$0.005**. A call that fails is never
charged.

| Tool call | On a subscription | Paid from the wallet |
| - | - | - |
| Succeeds | free | \$0.005 |
| Fails | free | free |

So a turn paid from the wallet that runs ten commands on the machine adds \$0.05
to the cost of its tokens.

## What each column counts

| Column | What it counts |
| - | - |
| **Input** | Tokens sent to the model that were not served from its cache: your message, the parts of the conversation the model has not seen before, tool results. |
| **Output** | Tokens the model writes: its answer, its tool calls, and its reasoning when the model reasons. |
| **Cached input** | Input the model has already seen in this conversation and reads back from its cache. An agent's turns resend the whole conversation, so after the first turn nearly all of the input is cached: in practice the cached-input price is the one that decides what a long session costs. |

## Working out a call

A turn's charge is the sum of its lines: each token count times its price, plus
its tool calls, rounded up to the next millionth of a dollar.

For example, one turn on `theo-spark-1.3` that reads 200,000 tokens, of which
194,000 come from the cache, writes 1,500, and runs three commands on the machine:

| Line | Tokens | Price per 1M | Charge |
| - | - | - | - |
| Input | 6,000 | \$4.00 | \$0.024 |
| Cached input | 194,000 | \$0.20 | \$0.0388 |
| Output | 1,500 | \$12.00 | \$0.018 |
| Tool calls | 3 | \$0.005 each | \$0.015 |
| **Total** | | | **\$0.0958** |

Every charge is listed with its token counts on **Transactions**; see
[Extra usage](https://docs.splox.io/account/extra-usage.md#where-the-spend-is-shown).

## Your own key

A run on your own credential (an endpoint of yours, see [Models](https://docs.splox.io/concepts/model.md))
is paid to your provider, at your provider's prices. It is not charged here and
does not touch the wallet.

> [!NOTE]
>
> Prices can change. A change applies to calls made after it; a charge already
> written never changes.

> [card] [Extra usage](https://docs.splox.io/account/extra-usage.md)
>
> The wallet these prices are charged to.

> [card] [Plans](https://docs.splox.io/account/plans.md)
>
> The windows a plan buys, before the wallet is used.


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