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

# Pricing Reference

> Full rate card for VLM Run Orion agents: fixed-price tools, token rates, and the per-span cost ledger.

This is the complete rate card for [Orion agent pricing](/pricing). For the value overview, cost components, and worked examples, start with the [Pricing](/pricing) page.

## How a run cost is calculated

The standard dollar cost is the sum of model-token usage and fixed-price tools. The service-tier multiplier is applied once at the run level:

```
model_cost = (
    uncached_input_tokens × input_rate
    + cached_input_tokens × cached_rate
    + output_tokens × output_rate
) ÷ 1,000,000

standard_cost_dollars = model_cost + fixed_tool_cost
cost_dollars = standard_cost_dollars × mode_multiplier
```

Where:

* **model\_cost** includes orchestration tokens and token-billed tool usage. Orchestration uses the rates for the `model` ID you selected; tool tokens are itemized on the per-span ledger
* **fixed\_tool\_cost** is the sum of fixed-price tools such as segmentation, generation, OCR, and sandbox execution
* **mode\_multiplier** is determined by the service tier: 1.0× standard, 0.5× flex, or 1.8× priority

Every run itemizes these components into a [per-span cost ledger](/pricing#the-per-span-cost-ledger), applying the multiplier once at the run level (`effective_cost = SUM(span.cost) × mode_multiplier`). The rates that feed each span are below.

## Token rates

A run has two kinds of token usage:

* **Orchestration tokens**: the Orion agent reasoning, dispatching tools, and writing the response. These bill at the rates for the `model` ID you selected.
* **Tool tokens**: the LLM-backed tools the agent calls. These are billed by the tokens each tool consumes, at the rate for the model that ran that tool (which may differ from the `model` ID you selected). The ledger itemizes each span.

The ledger is the source of truth for each charge: it records the token counts behind every span. These tools are token-billed rather than fixed-price:

| Tool                             | Category | Billing Basis                                                                                         |
| -------------------------------- | -------- | ----------------------------------------------------------------------------------------------------- |
| Caption & Tag                    | Image    | LLM tokens                                                                                            |
| Detect, Point                    | Image    | LLM tokens                                                                                            |
| UI Parsing                       | Image    | **Free**                                                                                              |
| Document Parsing / Extract / VQA | Document | LLM tokens (+ \$0.001/page when `grounding` or `confidence` is enabled on Orion-2 `document.extract`) |
| Caption, Summary & Transcribe    | Video    | LLM tokens                                                                                            |

Quality tiers (`fast` / `auto` / `pro`) use the tier rates below. Pinned Orion-2 backbones use model-specific input/output rates. The full list of accepted model IDs is in the [Chat Completions API reference](/api-reference/v1/post-chat-completions).

### Orion 2 model token rates

Rates are USD per 1M tokens. Orchestration tokens use the rates for the `model` ID you selected. Tool-token spans are itemized on the ledger. Cached input tokens are billed at the listed fraction of the input rate.

#### Quality tiers

Curated Orion-2 tiers. Fast and Auto share the same token rates; Pro is higher.

<table align="center">
  <thead>
    <tr>
      <th align="left">Model ID</th>
      <th align="center">Input</th>
      <th align="center">Output</th>
      <th align="center">Cached input</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td><code>vlmrun-orion-2:fast</code></td>
      <td align="center">\$0.30</td>
      <td align="center">\$2.50</td>
      <td align="center">10% of input</td>
    </tr>

    <tr>
      <td><code>vlmrun-orion-2:auto</code></td>
      <td align="center">\$0.30</td>
      <td align="center">\$2.50</td>
      <td align="center">10% of input</td>
    </tr>

    <tr>
      <td><code>vlmrun-orion-2:pro</code></td>
      <td align="center">\$1.00</td>
      <td align="center">\$10.00</td>
      <td align="center">10% of input</td>
    </tr>
  </tbody>
</table>

#### Open-weight pins

These open-weight Orion-2 backbones bill at the Fast / Auto token rates:

<table align="center">
  <thead>
    <tr>
      <th align="left">Model ID</th>
      <th align="center">Input</th>
      <th align="center">Output</th>
      <th align="center">Cached input</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td><code>vlmrun-orion-2:qwen3.6-35b-a3b</code></td>
      <td align="center">\$0.30</td>
      <td align="center">\$2.50</td>
      <td align="center">10% of input</td>
    </tr>

    <tr>
      <td><code>vlmrun-orion-2:gemma4-26b-a4b</code></td>
      <td align="center">\$0.30</td>
      <td align="center">\$2.50</td>
      <td align="center">10% of input</td>
    </tr>

    <tr>
      <td><code>vlmrun-orion-2:cosmos3-nano</code></td>
      <td align="center">\$0.30</td>
      <td align="center">\$2.50</td>
      <td align="center">10% of input</td>
    </tr>
  </tbody>
</table>

#### Pinned hosted backbones

These variants bill at model-specific pass-through rates (not the Fast / Auto / Pro tier rates):

<table align="center">
  <thead>
    <tr>
      <th align="left">Model ID</th>
      <th align="center">Input</th>
      <th align="center">Output</th>
      <th align="center">Cached input</th>
      <th align="center">Thinking</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td><code>vlmrun-orion-2:kimi-2.6</code></td>
      <td align="center">\$0.66</td>
      <td align="center">\$3.41</td>
      <td align="center">22% of input</td>
      <td align="center">N/A</td>
    </tr>

    <tr>
      <td><code>vlmrun-orion-2:muse-spark-1.1</code></td>
      <td align="center">\$1.25</td>
      <td align="center">\$4.25</td>
      <td align="center">10% of input</td>
      <td align="center">N/A</td>
    </tr>

    <tr>
      <td><code>vlmrun-orion-2:gemini-flash-3.6</code></td>
      <td align="center">\$1.50</td>
      <td align="center">\$7.50</td>
      <td align="center">10% of input</td>
      <td align="center">\$7.50</td>
    </tr>

    <tr>
      <td><code>vlmrun-orion-2:grok-4.5</code></td>
      <td align="center">\$2.00</td>
      <td align="center">\$6.00</td>
      <td align="center">25% of input</td>
      <td align="center">N/A</td>
    </tr>

    <tr>
      <td><code>vlmrun-orion-2:opus-4.8</code></td>
      <td align="center">\$5.00</td>
      <td align="center">\$25.00</td>
      <td align="center">10% of input</td>
      <td align="center">N/A</td>
    </tr>

    <tr>
      <td><code>vlmrun-orion-2:gpt-5.5</code></td>
      <td align="center">\$5.00</td>
      <td align="center">\$30.00</td>
      <td align="center">10% of input</td>
      <td align="center">N/A</td>
    </tr>
  </tbody>
</table>

<Note>
  Cached input is billed at the listed fraction of the input rate (for example, 10% of input means cached tokens cost \$0.10 when input is \$1.00 per 1M). Thinking tokens, when present, bill at the Thinking rate. Costs scale with tokens consumed, so simple inputs are cheaper than complex ones. See [model variants](/agents/code-execution#model-variants) for when to pin each backbone.
</Note>

### LLM orchestration

Every agent run includes LLM token costs for reasoning, tool dispatch, and response synthesis, charged at the selected `model` ID's input/output rates. Pinned frontier backbones (for example `vlmrun-orion-2:gpt-5.5` or `vlmrun-orion-2:opus-4.8`) use their own model-specific rates from the table above. A typical single-tool Fast or Auto run adds approximately \$0.01–\$0.03 (or \$0.03–\$0.08 on Pro) in orchestration costs. Multi-step and frontier-model runs will be higher.

## Fixed-price tools

These tools have a fixed dollar price per unit.

### Image

Priced **per image**. Fast and Auto share the same rates; Pro is higher where noted.

<table align="center">
  <thead>
    <tr>
      <th align="left" width="24%" style={{ whiteSpace: 'nowrap' }}>Capability</th>
      <th width="36%">Description</th>
      <th width="20%" align="center">Fast / Auto</th>
      <th width="20%" align="center">Pro</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>Segment</td>
      <td>Object segmentation (GPU-backed)</td>
      <td align="center">\$0.01</td>
      <td align="center">\$0.02</td>
    </tr>

    <tr>
      <td>Generate & Edit</td>
      <td>Generate or edit images from prompts</td>
      <td align="center">\$0.04</td>
      <td align="center">\$0.24</td>
    </tr>
  </tbody>
</table>

### Document

Priced **per page**.

<table align="center">
  <thead>
    <tr>
      <th align="left" width="24%" style={{ whiteSpace: 'nowrap' }}>Capability</th>
      <th width="36%">Description</th>
      <th width="20%" align="center">Fast / Auto</th>
      <th width="20%" align="center">Pro</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>OCR & Layout</td>
      <td>Extract text and detect page structure</td>
      <td align="center">\$0.01</td>
      <td align="center">\$0.04</td>
    </tr>

    <tr>
      <td>Extract grounding / confidence</td>
      <td>Orion-2 <code>document.extract</code> surcharge when <code>grounding</code> or <code>confidence</code> is enabled</td>
      <td align="center">\$0.001</td>
      <td align="center">\$0.001</td>
    </tr>
  </tbody>
</table>

### Video

Priced **per second** (generated video).

<table align="center">
  <thead>
    <tr>
      <th align="left" width="24%" style={{ whiteSpace: 'nowrap' }}>Capability</th>
      <th width="36%">Description</th>
      <th width="20%" align="center">Fast / Auto</th>
      <th width="20%" align="center">Pro</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>Generate & Edit</td>
      <td>Generate video from text/image prompts</td>
      <td align="center">\$0.15/sec</td>
      <td align="center">\$0.40/sec</td>
    </tr>
  </tbody>
</table>

<Note>
  Standard video generation produces \~6 seconds. At Fast / Auto pricing, \$0.15/sec × 6s = \$0.90; at Pro, \$0.40/sec × 6s = \$2.40.
</Note>

### Compute & execution

| Item                               | Fixed Price  |
| ---------------------------------- | ------------ |
| `compute.execution` (code sandbox) | \$0.001/call |

## Legacy pricing (`vlm-1`)

<Note>
  VLM Run is transitioning to dollar and token based pricing for Orion 2. The `vlm-1` model family and older per-request rates below are kept for reference during the transition and may be removed in a future update. New workloads should use Orion 2 pricing above.
</Note>

<Accordion title="vlm-1 pricing">
  The `vlm-1` model family is available on `/v1/image/generate` and `/v1/document/generate` routes with three quality tiers:

  | Model ID     | Tier           | Detail Level |
  | ------------ | -------------- | ------------ |
  | `vlm-1:fast` | Fast           | `lo`         |
  | `vlm-1:auto` | Auto (default) | `auto`       |
  | `vlm-1:pro`  | Pro            | `hi`         |

  Domain API token rates are in USD per 1M tokens:

  | Model ID     | Input  | Output | Cache Discount | Thinking |
  | ------------ | ------ | ------ | -------------- | -------- |
  | `vlm-1:fast` | \$0.25 | \$1.50 | 10%            | N/A      |
  | `vlm-1:auto` | \$0.26 | \$1.56 | 10%            | N/A      |
  | `vlm-1:pro`  | \$1.50 | \$7.50 | 10%            | \$7.50   |

  Set the model via `request.model`:

  ```python theme={"theme":{"light":"github-light","dark":"dark-plus"}}
  response = client.document.generate(
      file="invoice.pdf",
      domain="document.invoice",
      model="vlm-1:auto",
  )
  ```
</Accordion>

## FAQ

<AccordionGroup>
  <Accordion title="What's the difference between fixed-price and token-billed tools?">
    **Fixed-price** tools (segmentation, image generation, video generation, OCR, sandbox execution) have predictable dollar prices per unit. **Token-billed** tools (captioning, detection, document parsing, video analysis) bill by the tokens they consume, itemized on the per-span ledger.
  </Accordion>

  <Accordion title="How are pinned Orion-2 models priced?">
    Open-weight pins (`vlmrun-orion-2:qwen3.6-35b-a3b`, `vlmrun-orion-2:gemma4-26b-a4b`, `vlmrun-orion-2:cosmos3-nano`) bill at the Fast / Auto token rates (\$0.30 input / \$2.50 output per 1M). Hosted frontier pins such as `vlmrun-orion-2:kimi-2.6`, `vlmrun-orion-2:gpt-5.5`, `vlmrun-orion-2:opus-4.8`, `vlmrun-orion-2:muse-spark-1.1`, `vlmrun-orion-2:grok-4.5`, and `vlmrun-orion-2:gemini-flash-3.6` bill at their model-specific input/output rates in the [Orion 2 model token rates](#orion-2-model-token-rates) section. Fixed-price tools still use the Fast / Auto / Pro tool rates based on the resolved quality tier.
  </Accordion>

  <Accordion title="How is the total run cost determined?">
    It's the sum of model token costs (orchestration + token-billed tools) and fixed-price tool charges. The response includes `cost_dollars` (effective amount after the service-tier multiplier) and `standard_cost_dollars` (baseline).
  </Accordion>

  <Accordion title="What is the per-span cost ledger?">
    Every billable action during a run is recorded as a span: each LLM turn, each sandbox execution, and each tool call. The total is the sum of spans, with the service-tier multiplier applied once at the run level. This gives you full visibility into cost drivers.
  </Accordion>

  <Accordion title="Are utility tools free?">
    Yes. Image I/O, rotation, cropping, video trimming, document navigation, UI detection, and similar utilities are free.
  </Accordion>
</AccordionGroup>