Urgent.News

One page, thousands of outlets. See who else covered it.

Editions

AI

South Korea Joins U.S., China in Top Tier of Global AI Performance

South Korea’s national project to develop an independent artificial intelligence (AI) foundation model has been narrowed to three teams. Models participating in the evaluation ranked among the global top tier, behind only those from the United States and China, highlighting South Korea’s growing AI

South Korea has joined the United States and China in the top tier of global artificial intelligence (AI) performance, according to a recent evaluation. The Ministry of Science and ICT (MSIT) announced the results of its "Independent AI Foundation Model" project on August 18 at the Government Complex in Seoul. The evaluation involved three teams: Upstage, SK Telecom, and LG AI Research, all of which ranked among the global leaders in AI models, behind only those from the US and China.

The assessment considered benchmark scores, accounting for 40% of the total, along with expert assessments at 35% and user evaluations at 25%. Notably, all participating models achieved a minimum score of 31 on the Artificial Analysis Intelligence Index (AAII), a standard set by global AI model evaluation firm Artificial Analysis. South Korea's model was the only one from a country other than the US and China to achieve a score above 40 on this index.

Among the advancing teams, Upstage's "Solar Open 2" stands out with its ability to process up to 1 million tokens in context, comparable to the long-context processing capabilities of leading global models like OpenAI's GPT-5.6 Sol. SK Telecom's "A.X K2" exhibited exceptional performance in mathematical reasoning and Korean language capabilities, achieving a joint-best global accuracy score of 97.1% on a benchmark based on the American Invitational Mathematics Examination (AIME).

Additionally, LG AI Research's "K-EXAONE 2.0," with 750 billion parameters, was the largest AI model developed in South Korea and ranked ninth globally for reducing hallucinations, a common issue for AI models.

Motif Technologies' "Motif 3," which was eliminated, still performed well, scoring 47 on the AAII and ranking 10th among corporate AI models worldwide. South Korea, alongside the US and China, is currently the only country with an AI model on Epoch AI's "Notable AI Models in 2026" list. The MSIT aims to bolster the development of these models by expanding support for Nvidia's B200, a high-performance graphics processing unit (GPU).

This support will increase from 768 units in the first half of the year to approximately 1,000 units in the second half to accelerate model development. The government is also exploring a new, unprecedented support system with related ministries to further enhance South Korea's global AI competitiveness, with specific details set to be announced soon.

Written by urgent.news from BusinessKorea's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at businesskorea.co.kr →

More in AI

Why I Chose Text Classification Tagging APIs: OpenAI, Claude, Gemini JSON Accuracy

Short answer: for an e-commerce backend that scores job candidates against a rubric, I would choose the simplest API that can hold a JSON contract, expose cost per tenant, and let the team change…

  • OpenAI, Claude, and Gemini offer strong classification models for job candidate data.
  • Performance varies with malformed input, impacting system SLO reliability.
  • Infrai provides portable, provider-agnostic API integration with consolidated billing.

More from Tuesday 18 August →