> ## 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 > Dollar and token based pricing for VLM Run Orion agents. VLM Run Orion agents are priced in **US dollars**. Every response tells you exactly what a run cost. ## What makes up your cost Every run's cost comes from two components: * **Fixed-price tools:** GPU and API actions charged at a flat rate per unit. * Examples: image generation and editing, OCR and layout, and video generation and editing. * **Token-billed model usage:** Orion reasoning and LLM-backed tools charged by input and output tokens. * Examples: image captioning, document extraction, and video analysis. A single run typically combines both: the agent reasons (tokens), calls a tool or two, and returns a result with the total attached. ## How a run adds up Your cost is the sum of the two components, then scaled by your [service tier](#service-tiers): ``` run cost = token-billed usage + fixed-price tools total cost = run cost × service-tier multiplier ``` ## The per-span cost ledger Every Orion 2 run returns a **per-span cost ledger**: a line item for each billable action, priced as either a fixed-price tool or token-billed usage. This is where the two components become concrete, so you can see exactly what drove a run's cost. | Span Type | Priced By | Example | | --------------------- | -------------------------------------- | ------------------------------------------------- | | `mllm` | Model tokens (input + cached + output) | One LLM reasoning turn | | `execute_code` | Fixed price (\$0.001/call) | One sandbox code execution | | `tool` (fixed-price) | Fixed dollar price | `vlmrun.image.generate`, `vlmrun.image.segment` | | `tool` (token-billed) | Model tokens consumed | `vlmrun.document.extract`, `vlmrun.video.caption` | The multiplier is applied once, at the run level (`effective_cost = SUM(span.cost) × multiplier`), and the response hands you the totals directly, so you never have to add up spans yourself: | Response Field | What It Tells You | | ----------------------- | -------------------------------------------------------- | | `cost_dollars` | What you actually pay, after the service-tier multiplier | | `standard_cost_dollars` | Baseline cost at the standard tier | | `savings_dollars` | Amount saved versus standard (positive on `flex`) | | `service_tier` | Which delivery tier ran the request | ### An example ledger Here is an illustrative, ballpark example for a single document processed by Orion 2 Auto in [program mode](/agents/code-execution). Actual token usage and cost vary with the document and the requested output. | Span | Billed Usage | Cost | | --------------------------- | ------------------------------------- | ------------ | | `execute_code` | Fixed price | \$0.0010 | | `vlmrun.document.extract` | 6,000 input + 1,000 output tokens | \$0.0020 | | `vlmrun.document.grounding` | Fixed price | \$0.0010 | | **Total** | **2 fixed-price spans, 7,000 tokens** | **\$0.0040** | Running the generated program contributes the fixed `execute_code` charge, extraction is billed on the tokens it consumed, and grounding adds a second fixed charge. The run total is the sum of the three spans, before the service-tier multiplier is applied. For the full rate card behind each span, see the [Pricing Reference](/pricing/reference). ## Choose your tiers Two independent dials control quality and delivery. Quality picks the models; the service tier scales the whole bill. ### Model quality Use Fast, Auto, or Pro as a curated default, or pass a specific model ID to pin a backbone. Token rates vary by model; see the [Pricing Reference](/pricing/reference#orion-2-model-token-rates) for the full rate card. | Tier | Model ID | When to Use | | -------- | --------------------- | ------------------------------------------------------------------------ | | **Fast** | `vlmrun-orion-2:fast` | Speed and cost-efficiency. Lighter models, fastest processing. | | **Auto** | `vlmrun-orion-2:auto` | Recommended default. Uses Fast tools, upgrades to Pro for complex tasks. | | **Pro** | `vlmrun-orion-2:pro` | Maximum quality. More powerful models, higher cost, longer processing. | VLM Run hosts additional Orion-2 backbones (open-weight and frontier models) beyond these three tiers. Pass any supported `model` ID on [chat completions](/api-reference/v1/post-chat-completions) or agent execute to pin a backbone. Per-model input/output rates are in the [Pricing Reference](/pricing/reference#orion-2-model-token-rates). ### Service tiers Same models, same tools, same output: the service tier only changes how quickly the result comes back. Batches, backfills, and eval sweeps that nobody is waiting on cost half as much at `flex`; save the 1.8× `priority` premium for when a person is waiting on the response.
Tier Multiplier Cost Effect Best For
standard (default) 1.0× Base USD price Most production workloads.
flex 0.5× 50% of standard Background jobs with no one waiting: overnight batches, backfills.
priority 1.8× 180% of standard Interactive or blocking workflows where latency is user-visible.
A \$1.00 standard run costs \$0.50 at `flex` and \$1.80 at `priority`. Set it per request: ```python Python theme={"theme":{"light":"github-light","dark":"dark-plus"}} # Flex tier: 50% discount on total agent cost response = client.agent.execute( name="my-orion-agent", inputs={...}, service_tier="flex", ) # response.usage.cost_dollars → effective cost # response.usage.savings_dollars → amount saved ``` ```javascript Node.js theme={"theme":{"light":"github-light","dark":"dark-plus"}} // Priority tier: 1.8x premium, lowest latency const response = await client.agent.execute({ name: "my-orion-agent", inputs: {...}, serviceTier: "priority", }); ``` ```bash cURL theme={"theme":{"light":"github-light","dark":"dark-plus"}} curl -X POST https://api.vlm.run/v1/agent/execute \ -H "Authorization: Bearer $VLMRUN_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "name": "my-orion-agent", "inputs": {...}, "service_tier": "flex" }' ``` ## What a run actually costs See the following worked examples. The `cost_dollars` field in the response is always the exact figure. | Component | Type | Auto | | -------------------------- | ---------------- | ------------ | | Segment objects in 1 image | Fixed-price (1×) | \$0.01 | | Generate 1 image | Fixed-price (1×) | \$0.04 | | LLM orchestration | Tokens (approx.) | \~\$0.01 | | **Estimated total** | | **\~\$0.06** | Fixed-price tools dominate; token orchestration is a small share. | Component | Type | Standard Cost | | ------------------------------- | ------------------------ | -------------------- | | OCR & layout (50 pp) | Fixed-price: \$0.01/page | \$0.50 | | Parse / extract data | Token-billed | Varies by complexity | | LLM orchestration | Tokens (approx.) | \~\$0.02–\$0.05 | | **Estimated standard subtotal** | | **\~\$0.52–\$0.55** | | Flex service tier | 0.5× multiplier | × 0.5 | | **Estimated total with Flex** | | **\~\$0.26–\$0.28** | ## Full rates Every fixed-price tool and per-model token rate, plus the per-span ledger and cost formula. Per-token rates for VLM Run Gateway models, metered on every request.