SEOUL, South Korea, June 22, 2026 /PRNewswire/ -- Korea Deep Learning, a pioneer in enterprise-grade Document AI and vision intelligence solutions, today announced that its proprietary Document AI has secured the #1 overall ranking on ParseBench, a premier global document parsing benchmark organized by LlamaIndex.
ParseBench evaluates AI models on their ability to read, structure, and extract information from approximately 2,000 pages of complex, real-world enterprise documents across finance, insurance, and government sectors. In the latest evaluation, Korea Deep Learning ranked #1 overall in the VLM category with a score of 76.4, ahead of the next-highest-scoring models (75.0 and 67.8). The company also secured the top spot in Visual Grounding with a score of 78.8, establishing a commanding 19-point lead over the next-highest-scoring model (59.8).
The win was driven by KDL-Frontier-Parser-nano, a 1.2-billion-parameter single model. Despite its compact size, it delivered superior accuracy against massive Big Tech systems at a fraction of the operational cost, proving that specialized AI tailored for business workflows can outperform broad, general-purpose LLMs. Its core edge lies in Korea Deep Learning's Near-Zero Hallucination technology, which simultaneously analyzes layout, field relationships, and spatial positioning to minimize error rates and extract data with context-aware precision.
Trained on over 400 million images and documents, Korea Deep Learning's solutions support secure on-premise and SaaS deployments. Validated across major enterprise and public sectors, the technology powers administrative data restructuring for the Gyeonggi Provincial Government and serves over 80 major corporations, including Hyundai Capital and LG CNS—slashing document-processing times by up to 96%.
Today's ParseBench victory builds on a consecutive string of global milestones. In March 2026, the company's proprietary Vision-Language Model, KDL Frontier, clinched the #1 spot in the English category of OCRBench v2 with a score of 68.1, outperforming other leading models. In recognition of this sustained technological leadership, Forbes Asia officially named CEO Jihyun Kim to its prestigious 2026 '30 Under 30' list in the Consumer & Enterprise Technology category last month.
"Our strategy has always been to build specialized AI that decodes complex document structures as intuitively as a human, rather than focusing on massive LLMs," said Jihyun Kim, CEO of Korea Deep Learning. "This milestone validates our architecture as the most practical solution for the 'AI Agent' era, enabling autonomous decision-making and total workflow automation."
Backed by an $8 million (KRW 12 billion) Series A funding round secured in December 2025 and a strategic partnership with NVIDIA, Korea Deep Learning is aggressively accelerating its global expansion into international enterprise markets.
About Korea Deep Learning
Founded in 2019, Korea Deep Learning is a premier vision intelligence and Document AI company developing enterprise-grade AI agents that automate document-based workflows. Its proprietary DEEP Agent platform combines advanced OCR, document parsing, and vision-language models to help organizations seamlessly extract, structure, and act on information from complex documents, images, and visual data.
** This press release is distributed by PR Newswire through automated distribution system, for which the client assumes full responsibility. **
Korea Deep Learning Secures Back-to-Back Global AI Benchmark Wins Over Big Tech
