Upstage Unveils 'Solar Pro 4' with Developer Platform Listings

by Shin Hye An Posted : August 14, 2026, 08:48Updated : August 14, 2026, 08:48


Upstage announced the release of its self-developed large language model (LLM) 'Solar Pro 4' on August 14.

Solar Pro 4 is a next-generation commercial artificial intelligence (AI) model that enhances reasoning capabilities and agent performance compared to its predecessor, 'Solar Pro 3.' It is optimized for stable execution of workflows required in real work environments, such as document understanding, information extraction, long-context reasoning, and continuous decision-making, rather than focusing solely on single-answer benchmarks.

According to Upstage, immediately after its release, Solar Pro 4 was listed alongside the AI agent 'Hermes Agent' developed by NUS Research and the API routing platform 'OpenRouter' on global developer platforms. Upstage is the first Korean company to be officially listed as a provider among major U.S. tech firms like OpenAI, Anthropic, Google, and NVIDIA, as well as leading Chinese companies like DeepMind, Moonshot, and Minimax.

Upstage stated, "This is the first time a Korean model has been recognized, establishing a foundation for expanding the reach of Korea's sovereign AI on a global scale."

Within just three days of being listed on OpenRouter, Solar Pro 4 surpassed 80 billion tokens in cumulative token consumption. Additionally, it became the first Korean model to enter the 'Agent Arena' leaderboard, which compares agent performance among major global models. The company noted that it ranked alongside NVIDIA's 'NemoTron 3 Ultra' in terms of performance, breaking into a ranking traditionally dominated by U.S. and Chinese models.

The improvements focus on critical areas that determine the success of agent tasks. In the 'Terminal-Bench v2.1,' which evaluates the ability to complete multi-step tasks in a terminal, Solar Pro 4 scored 57 points, an increase of 4.8 times compared to its predecessor. In the 'Tau3-Banking' benchmark, which assesses tool usage during conversations, it achieved a score of 23, a 2.6-fold improvement.

The model's ability to handle long and complex documents without failure has also emerged as a key strength. Solar Pro 4 scored 71 points in the long document understanding benchmark (AA-LCR), demonstrating over 2.3 times better performance than its predecessor. The reduction in failures to maintain context within documents has improved both accuracy and token consumption, leading to lower operational costs.

Furthermore, the Korean language performance and document processing capabilities tailored for domestic businesses and public institutions have significantly improved. Solar Pro 4 is integrated into Upstage's agent document processing solution, 'Upstage Studio,' enabling it to extract information from various document formats, including Korean documents, and perform tasks such as summarization, analysis, and translation simultaneously.

Overall performance is rated at 42 points by the global AI evaluation agency 'Artificial Analysis,' placing it in the same category as general-purpose frontier models. This score represents an improvement of more than three times compared to its predecessor. It surpassed big tech models like NVIDIA's 'NemoTron 3 Ultra' (38 points) and Google's 'Gemini 3.5 Flash-Light' (37 points), as well as competing sovereign models like Mistral's 'Medium 3.5' (30 points) and Cohere's 'Command A+' (23 points).

Kim Seong-hoon, CEO of Upstage, stated, "What AI needs in practice is not just the ability to perform well on tests, but reliability in completing assigned tasks. The fact that Solar Pro 4 is the first Korean model to be listed on global developer platforms is evidence that our AI technology is effective in real-world applications. We will continue to develop more practical AI based on proven technology in the field."

Since the release of 'Solar Pro 3' in March of this year, Upstage has been continuously enhancing the reasoning and agent performance of its commercial LLM. Subsequently, it applied its own reinforcement learning technology to improve the tool usage, coding, and terminal operations of Solar Pro 3. In April, it launched the agent document processing solution 'Upstage Studio,' expanding the application range of its AI model in corporate tasks.





* This article has been translated by AI.