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

# Pricing and limits

> Token-based pricing per model role, the per-request minimum, and the rate and size limits.

Billing is per token. Two models do the work — `sdvm-audit-1` produces every verdict, including every vote and the reaudit inside a refine, and `sdvm-fix-1` produces every fix — and each token bills at the rate of the model that consumed it.

| Model          | Input                | Output               |
| -------------- | -------------------- | -------------------- |
| `sdvm-audit-1` | \$0.12 per 1M tokens | \$0.45 per 1M tokens |
| `sdvm-fix-1`   | \$0.12 per 1M tokens | \$0.45 per 1M tokens |

Minimum \$0.01 per request, rounded up to the whole cent. Every response reports `cost_usd`, `input_tokens` and `output_tokens`.

Rates can change. `GET https://api.sdvm.ai/pricing` returns the live rates and limits (no key needed). The table above is a snapshot from 2026-09-16, rounded up to the cent — the endpoint has the exact figures and is what a request is actually billed at.

## What drives the cost

* An audit's output tokens usually exceed its input tokens.
* `votes=N` audits each sample N times and bills the work actually done.
* `max_attempts=N` gives the fixer up to N tries; only flagged samples are tried again.
* `refine` runs `audit -> fix -> reaudit`; only samples the fix changed are re-audited.

## Limits

|                            | Limit                                                                                                   |
| -------------------------- | ------------------------------------------------------------------------------------------------------- |
| Samples per request        | 100                                                                                                     |
| Requests per minute        | 100 per API key                                                                                         |
| Votes per audit stage      | 1 to 9, odd                                                                                             |
| Fix attempts               | 1 to 5                                                                                                  |
| Sample size                | About 186,000 tokens (roughly 745,000 characters) per sample. The error message states the exact limit. |
| Estimated cost per request | \$5.00. A request whose pre-flight estimate exceeds it is rejected before any work is done.             |

Credits are bought on your [profile page](https://sdvm.ai/profile).