Recently, the South Korean artificial intelligence laboratory Upstage officially launched a commercial large model called Solar Pro4, tailored for complex agent tasks. This new model features a context window of up to 512K and supports an output length of up to 128K, allowing users to easily load and process multiple contracts, reports, and data files in a single session.

In terms of pricing, Solar Pro4 demonstrates highly competitive advantages. The input price per 1 million Tokens is $0.3, the cache read input price is $0.06, and the output price per 1 million Tokens is $1.2.

image.png

Focusing on performance for agent tasks, the model has been deeply optimized for real-world scenarios, with a particular emphasis on long documents, terminal tasks, and multi-turn tool calls. In official core tests, it scored 57 on Terminal-Bench v2.1, 23 on τ³-Banking, and 71 on AA-LCR. Upstage demonstrated this through a simulated task of analyzing store locations for a coffee brand: the model completed policy screening among 10 candidate sites using just three prompts and automatically generated three complete deliverables in sequence - an Excel workbook, a review report, and a presentation.

Currently, Solar Pro4 has fully debuted on several international authoritative AI model evaluation platforms. As the first model from a South Korean laboratory to appear on the Agent Arena, Code Arena: WebDev, and Text Arena rankings, it is ranked 44th on the Agent Arena list, matching the capabilities of well-known models such as MiniMax M2.7 and Nemotron3Ultra.