Upstage AI

Upstage AI

Document intelligence and Solar LLM APIs from Korea

Officially listed
English
Paid
AI基础设施
0favorites
0views
Promote this tool
Upstage AI screenshot

Tool details

Upstage is a Korean AI company founded in 2020, working on two fronts: document intelligence and its own large language models. It turns scans, PDFs and emails into structured data, then hands them to the Solar family for extraction, question answering and agent workflows, sold into insurance, healthcare and manufacturing pipelines.

Core capabilities

  • Document Parse: turns complex layouts and scans into model-readable structure, billed per page with automatic tier selection
  • Information Extract: pulls structured fields out of invoices, claims and contracts at standard or enhanced accuracy
  • Solar Pro 4: the flagship model for agent work, where context length and long-document reasoning are the selling points, in English, Korean and Japanese
  • Document OCR and classification: text extraction and document classification as separate APIs, suited to batch preprocessing before ingestion
  • Upstage Studio: chains parsing, extraction and classification into workflows without code, each agent shipping its own API endpoint
  • Private delivery: on-premises deployment plus procurement through AWS, Azure and Snowflake marketplaces

Where it fits Insurance claims, medical record sorting, manufacturing quality documents and financial KYC: processes where documents are many, layouts are messy and results must be auditable. What buyers want is stable extraction and reproducible accuracy, not a chat experience.

Why it stands out Its strength sits in document benchmarks rather than general intelligence: 80.51 on ParseBench in the vision-language category ranks first, at roughly one fifth of the per-page cost of leading closed models. The Solar family does not lead general reasoning leaderboards, so treating it as an all-round chat model will disappoint.

Editorial take A fit for companies that treat document processing as a production line: fine-grained APIs, a private deployment path and pricing down to the page. Individual developers will find the prepaid threshold high, and the text-only Solar Pro 4 cannot read images directly, so scans still need the parsing layer first.

Pricing

Subscriptionfrom $0.0015 per page (Document OCR)Each Studio agent includes 10 free runs; schools, hospitals and nonprofits can apply for up to one year of free allowance
Solar Pro 4
$0.30 per 1M input tokens

$0.06 cached input and $1.20 output; flagship agent model, with a phased discount schedule published on site

Solar Mini 4
$0.10 per 1M input tokens

$0.01 cached input and $0.40 output; 3B active parameters for high-frequency, low-cost calls

Document Parse
$0.01 per page

Standard tier; enhanced tier is $0.03 per page, with automatic per-page tier selection

Document OCR and classification
$0.0015 per page

OCR unit price; document classification is $0.004 per page

Information Extract
$0.04 per page

Standard tier; enhanced tier is $0.06 per page

Upstage Studio
from $0 per agent

Each agent includes 10 free runs, then bills as the sum of the steps it runs

Prepaid commitment (monthly)
from $100 per month

Explore tier adds 10% bonus credits; Build from $500 and Scale from $5,000; yearly billing saves 20%

Enterprise
Custom quote

Private deployment, dedicated rate limits and support, priced by contract

Additional notes

All list prices exclude 10% VAT; purchased credits are non-refundable once used; Solar Pro 3, Solar Pro 2, Solar Mini and the first-generation Embed were retired on 2026-10-30.

FAQ

There is no standing free tier; usage is metered. Document extraction starts at $0.0015 per page and language models are billed per 1M tokens. Each Studio agent includes 10 free runs, and schools or nonprofits can apply for free allowance.

User reviews

User Reviews

No reviews yet

No one has reviewed this tool yet