Upstage Solar is a family of large language models for developers building document analysis, coding, conversational, and business workflow applications. Solar Pro 4, the model described here, provides AI Model Serving & Inference through Upstage’s hosted API. It generates text, returns structured data, and selects application-defined tools for workflows that require information retrieval or external actions.

Key Capabilities

Long-context processing: Solar Pro 4 supports a 512K-token context window and up to 128K output tokens. Applications can supply lengthy documents and conversation histories for summarization, comparison, question answering, and report drafting.
Configurable reasoning: Developers set reasoning effort for tasks such as planning, mathematical analysis, and debugging. When enabled, reasoning is returned separately from the final answer, with its token usage reported by the API.
Tool calling: Applications describe functions and their arguments, and Solar returns calls for the application to execute. Parallel tool calling allows multiple independent requests in one response. Tool results can be supplied back to the model to continue the workflow.
Structured outputs: Responses can follow a supplied JSON Schema, supporting extraction, classification, and other workflows that require defined fields and types. Streaming delivers output incrementally.
Multilingual generation: Solar Pro 4 supports English, Korean, and Japanese input and output.

Audience & Use Cases

Solar serves application developers and enterprise AI teams building document assistants, support applications, coding workflows, and agents that interact with business systems. Document-based uses include comparing contracts, synthesizing reports, and extracting information into structured records. Tool execution and workflow control are supplied by the surrounding application or agent runtime.

Technical Specifications

Solar Pro 4 supports OpenAI-compatible Chat Completions and Responses APIs, with browser-based testing through Upstage’s Playground. Hosted usage is metered by input, cached-input, and output tokens. Upstage also offers an on-premises deployment option by arrangement. Its documentation states that API inputs are not used for model training.

Upstage Solar

What is Upstage Solar?

Categories & Use Cases