Upstage: Solar Pro 3
upstage's efficiency model. Context window: 131K tokens.
Scores by test
Methodology →What you need to know
Solar Pro 3 is optimized for high-precision formatting and global accessibility, achieving perfect internal scores for structured output and multilingual capabilities. Its strength in strategic analysis and constrained rewriting makes it a reliable choice for developers needing strict adherence to schema or translation-heavy workflows.
The model is priced competitively at a blended cost of $0.487 per million tokens, offering a low-cost entry point for high-context tasks given its 131K window. However, this value is offset by significant performance gaps in specific data types; it is nearly incapable of handling tabular data and shows a tendency to over-refuse benign requests due to poor safety calibration.
Technical performance is inconsistent across complex reasoning tasks. While it handles tool calling and faithfulness well, it struggles with scientific coding, as evidenced by a 24.7% SciCode score, and shows mediocre results in agentic planning and creative problem solving.
Use this model if your primary requirement is reliable JSON/structured output or multilingual support at a low price point. Skip this model if your application requires tabular data processing, high-level scientific coding, or a nuanced safety filter that avoids over-refusal.
Strengths — Top 3
Relative weaknesses — Bottom 3
Similar models