LG Releases Korea's Largest AI Model 'K-EXAONE 2.0' as Open Source

*Editor's Note: 'AI PRISM' (Personalized Report & Insight Summarizing Media) is an "artificial intelligence (AI)-based personalized news recommendation and summarization service" developed with support from the Korea Press Foundation. It selects and provides six customized news items by reader type.

1. K-EXAONE Triples in Scale, Surpasses China's Leading AI in Performance

- Key Summary: LG AI Research has released 'K-EXAONE 2.0,' a model with 750 billion parameters, on the open-source platform Hugging Face. This is more than three times larger than the first model, which had 236 billion parameters, making it the largest AI foundation model unveiled in Korea. In particular, it scored 94.4 on the long-text comprehension evaluation index (OpenAI-MRCR), far surpassing the 71.5 scored by China's Zhipu AI 'GLM-5.1,' and it applies the Apache 2.0 open-source license, which allows free commercial use. LG AI Research also said it plans to unveil an industry-specialized AI model next week.

2. "Turning Lab Achievements into Growth Engines": DGIST Technology Holdings Launches Full-Scale Investment

- Key Summary: DGIST Technology Holdings, a subsidiary of DGIST, has formed its first investment fund of about 2 billion won since incorporation. The fund plans to invest in startups founded by DGIST faculty and researchers in national strategic technology fields such as AI, robotics, semiconductors, and bio-healthcare, as well as early-stage deep tech companies. It will also operate a "growth acceleration network" linked with major domestic venture capital firms (VCs) to support follow-on investment, and plans to gradually expand the fund's size by utilizing policy investment programs such as the government's fund of funds. In addition, it plans to link with TIPS (Tech Incubator Program for Startup), a private investment-led technology startup support program, to help portfolio companies win TIPS selection.

3. Despite 23x Less Private AI Investment, China Chases U.S. to 2.7% Gap Through 'All-Out National Effort'

- Key Summary: In evaluations on the large language model (LLM) assessment platform Arena Leaderboard, the performance gap between China's and the U.S.'s top-tier AI models has narrowed to 2.7%. Anthropic's 'Claude Opus 4.6' led with 1,503 points, narrowly ahead of China's ByteDance 'Doubao-Seed 2.0' (1,464 points). On the other hand, U.S. private AI investment last year reached $285.9 billion, 23 times that of China ($12.4 billion), and the U.S. also led China by more than 10-fold in the number of data centers and the number of AI companies attracting new investment. However, China is ahead of the U.S. in industrial robot adoption and the number of AI doctoral graduates, narrowing the gap through its state-led strategy.

4. Different from Last Year: IBK Investment & Securities Grows Its Presence in the IPO Market

- Key Summary: IBK Investment & Securities has grown its presence in the initial public offering (IPO) market, filing seven KOSDAQ listing preliminary reviews in the first half of this year and winning approval for three of them. This is a significant increase compared to last year, when it managed only one listing for Korea PIM, and marks its highest-ever performance level. Of five new listings and two SPAC (special purpose acquisition company) absorption merger listings, it served as sole underwriter for most, excluding HLB Genomics, which listed on KOSDAQ on the 24th of this month. It is also the only firm to file a KONEX listing, expanding its support to early-stage and small and medium-sized companies with low access to capital markets.