Upstage was formed in 2020, in South Korea and has developed its own large language models (LLMs) like Solar LLM, which are used along with advanced Document AI Solutions to help organizations improve their workflows. As an organization that uses artificial intelligence to improve workflow efficiency in various fields (healthcare, insurance etc.) they have made it their mission to use artificial intelligence in a way that makes it available and valuable to businesses around the globe. As of this time, Upstage has established itself as a leading provider of high performance AI, and is led by professionals who are experienced in the field of AI, having worked for Naver Clova and other major technology firms. They employ over 100 people in offices located in Seoul, San Francisco, and Tokyo.
Founded by CEO Sung Kim (former Naver Clova), CTO Stan Lee, and CSO Lucy Park, bringing together the combined experience of AI experts from Naver, Meta, Amazon, and many other major technology companies.
Received interest from more than 100 companies in the finance, education, IT, and manufacturing sectors during the course of their first year.
Reached two years in business with offices in Seoul, San Francisco, and Tokyo, and a focus on developing Document AI and LLM solutions.
Recognized as one of the top 100 most promising AI companies in the world in terms of innovation.
GPU servers with WEKA Data Platform for high-performance AI model training; available on AWS and on-premises
Proprietary Solar LLM family using Depth-Up Scaling (DUS) trained on 7B open-source models with focus on preventing hallucinations, data leak prevention, and enterprise-grade performance; Document AI with multimodal OCR capabilities achieving 95%+ accuracy
Based on company website, WEKA case study, and technical claims; specific frameworks from expert backgrounds (Naver Clova experience)
Upstage's offerings of Chat Completions (Solar LLM), Document Parsing and OCR for text extraction, Information Extraction for structured data from documents, Embeddings for semantic search, Document Classification, and Groundedness Check to verify the accuracy of the responses generated by the AI.
Upstage's offering of Solar Mini and Solar Pro 3 chat models. The Solar Mini is the default model for simple text generation, while the Solar Pro 3 is the preferred model for more complex text generation. Embedding models offered include embedding-query and embedding-passage variants.
Go to console.upstage.ai, make an account, get an api key under the "api Keys" menu, and then use that to login to your app. The upstage sdk can be found for python, javascript, and rust; or you could use curl/rest with an openai sdk for backwards compatability.
Yes. Upstage chat completions, as well as most of our other apis, have been implemented using the openai sdk so switching to either provider or using both interchangeably should be very simple.
Yes. We provide integration for both langchain and llamaindex which will allow you to use solar models along with the rest of the upstage api inside of those two popular ai frameworks.
OCR accepts pdf and other document types to perform text extraction. API endpoint is post/v1/document-digitization with model="ocr". Upstage also supports both synchronous and asynchronous document extraction for larger documents.
Send a document and optionally a schema to extract structured data based upon your needs. Upstage also provides a feature called schema generation with model="schema-generate" to automatically generate extraction schemas for your needs. Both synchronous and asynchronous endpoints are provided depending upon the size of the document being processed.
Upstage embeddings convert text into vectors for searching similar content, comparing vectors for similarity, performing vector clustering, and doing vector-based classification. We offer dual models optimized for creating embeddings for queries with embedding-query and document embeddings with embedding-passage.
Upstage has API documentation and examples available for testing. You may create a free account at console.upstage.ai to test out the APIs. There is no specific sandbox option mentioned, but you may use your dev API key for testing prior to deploying to production.
Never share your api key publicly. Store it securely, we recommend environment variables, and upstage provides complete documentation on how to properly secure your authentication. For enterprise level security, please contact us through sales for custom options.
Upstage presents itself to be an AI-specific firm that provides enterprise-level Large Language Model (LLM) & Document AI capabilities via easy-to-use API's. The Solar Family of Models offer competitive performance with strong OpenAI Compatibility, which will allow for a seamless Integration into existing systems. In addition to its robust LLM capabilities, Upstage has created a Complete Document Processing Suite (OCR, Parsing, Extraction, Classification), which separates it from other General-Purpose LLM Providers. This makes it a particularly attractive option for enterprises that have large amounts of documents they process.
UpStage Is Best Suited For Enterprise Customers Who Require Production Ready Document AI Combined With Competitive LLM Capabilities. UpStage Provides Significant Value To Document Heavy Workflows Where Competitors Don't Specialize.
Upstage is an Enterprise-Grade provider of AI focused specifically on Document Processing and Large Language Model (LLM) capabilities. The company offers a wide range of APIs that support Chat functionality, Document Parsing, Optical Character Recognition (OCR), Information Extraction, Embeddings, and Classification. A strong SDK ecosystem supports LangChain, LlamaIndex, and Portkey, making it easier for companies to integrate the service into their systems. Additionally, well-documented APIs provide OpenAI compatibility allowing companies that have used OpenAI before to quickly adapt to Upstage's interface.
Good — comprehensive technical documentation from official Upstage console, GitHub cookbook examples, and integration guides. API reference complete with code examples. Pricing and detailed SLA metrics not publicly available; enterprise information requires direct contact.
Upstage can be integrated with most major AI frameworks such as LangChain, LlamaIndex, Portkey, and AISDK. This will allow developers to replace existing models from other vendors with Upstage models, which may also help teams who wish to standardize on a particular framework.
Upstage has a GitHub Cookbook (https://github.com/UpstageAI/cookbook) that showcases example API calls and some innovative ways to utilize the service. Upstage also hosts a console at https://console.upstage.ai where users can manage their API keys, view documentation, and monitor how often they have called the API.
Upstage differentiates itself from competitors by offering advanced document intelligence APIs beyond traditional LLM capabilities. The suite includes but is not limited to; OCR, Document Parsing that serializes tables and figures, Information Extraction with Schema Generation, Document Classification, and Groundedness Verification for RAG Applications.
Solar Mini is the default lightweight model intended for simple applications. Solar Pro 3 is designed to handle more complex reasoning tasks. The embedding models are both Query and Passage Variants that are optimized based upon the specific application requirements of Vector Search and Similarity.
Recognized as a Frontier Model by Artificial Analysis, the 31B parameter LLM is the flagship model that has been shown to possess the best possible performance when it comes to complex reasoning, coding, and multilingual support — particularly in terms of its performance when processing the Korean language.
While still a high-performance model, the compact 10.7B parameter variant has been optimized for deployment in resource-constrained environments — making it a viable alternative to the 31B parameter version of the model.
A document processing solution that converts PDFs, scans, and other complex documents into easily readable, structured data using Optical Character Recognition (OCR), layout analysis, and chart/table extraction technology.
The complete toolbox to convert unorganized documents to organized data by using optical character recognition (OCR) and layout analysis to identify tables, images and contextually dependent information.
Extraction of precise data from contract, invoices and other commercial documentation with a confidence score.
Conversational platform without hallucination which allows users to load and communicate with their documents using natural language questions.
Unique training method to improve compact models to be competitive with large models such as GPT-4, allowing compact model to be deployed on a single GPU server at enterprise level.
Feature built to reduce hallucinations and errors while generating trustworthy AI answers with full audit trail.
Compact models are designed to run on each user's individual GPU server which reduces the cost of required infrastructure to maintain high performance.
Capability to connect models with customers' proprietary organization data and knowledge base.
Advanced language understanding in multiple languages including highly specialized performance in Korean.
Automation of claims processing, underwriting and document analysis
Processing of clinical documents and analysis of medical records.
Automated processing of contracts, invoices, and automated extraction of financial data from documents.
Automated processing of documents, forecast of demand, inventory optimization, and supplier management.
| Dimension | Upstage Solar Pro 2 | GPT-4o | LLaMA 4 Maverick |
|---|---|---|---|
| Parameter Count | 31 billion | Unknown (larger) | Unknown (larger) |
| Composite Score | 58 points | 41 points | 51 points |
| Deployment Model | Single GPU optimized | Proprietary API | Open-source option |
| Frontier Model Status | Recognized by Artificial Analysis | Established leader | Emerging competitor |
| Korean Language Performance | Exceptional | Standard | Standard |
