llmcloud.ai
Apps · Research & analysis

DocParse

Invoice and contract extraction at volume.

Tokens / 7d
29B
Routing directive
auto:vision
On our metal
44%
Licence
Proprietary

Document pipelines run the same prompt against millions of pages. Accuracy per dollar is everything, and a cheap vision model with a verification pass beats a frontier model on every page.

How it connects

  • Vision route with strict JSON schema per document type.
  • Cheap first pass, frontier escalation only on low-confidence pages.
  • Batch submission with relaxed latency SLAs.

Traffic pattern

Embarrassingly parallel, latency-insensitive, and the best fit for hosted open-weight vision models.

gemini-3.6-flashclaude-sonnet-4.5qwen3-vl-235b

Point it at the gateway

const res = await client.chat.completions.create({
  model: "auto:vision",
  response_format: { type: "json_schema", json_schema: invoiceSchema },
  messages: [{ role: "user", content: [
    { type: "text", text: "Extract line items." },
    { type: "image_url", image_url: { url: pageUrl } }] }] });