Skip to main content

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

On July 31, LG AI Research Institute quietly dropped a bombshell in the AI world: K-EXAONE 2.0, the largest artificial intelligence model ever built in South Korea, is now available on Hugging Face. This isn't just another incremental update—it's a bold statement of intent.

Developed under the Sovereign AI Foundation Model Project backed by Korea's Ministry of Science and ICT, K-EXAONE 2.0 packs a staggering 750 billion parameters, though only 37 billion are activated at any given time thanks to its hybrid attention MoE (Mixture of Experts) architecture. That's more than three times the size of its predecessor, making it a true giant in the field.

What's even more striking is that LG has chosen to fully open-source the model under the permissive Apache 2.0 license. In a country where AI development has often been guarded, this move is nothing short of revolutionary. It's a clear signal that Korea is ready to play in the big leagues of open-source AI.

Performance That Speaks Volumes

So, how does K-EXAONE 2.0 actually perform? The numbers are impressive. Across 24 benchmark tests, it averaged 70.1 points—a jump of over 10% from the first generation's 63.3. But the real fireworks are in coding and agent-based tasks, where performance improved by roughly 30%.

But LG isn't just looking at its own past; it's eyeing the competition. In long-context understanding, K-EXAONE 2.0 outshines Zhipu's GLM-5.1. It also surpasses both GLM-5.1 and Qwen3.5 in agent tool usage—a critical capability for real-world applications. When it comes to instruction following, it holds its own against heavyweights like GLM-5.1, DeepSeek V4 Pro Max, and Qwen3.5.

These aren't just random benchmarks; they're the metrics that matter for practical AI deployment. By going head-to-head with China's best open-source models, LG is positioning Korea as a serious contender in the global AI arena.

A New Triad in Open-Source AI

The release of K-EXAONE 2.0 comes at a time when the open-source AI landscape is heating up, especially with the buzz around Kimi K3. But this isn't just about joining the fray—it's about reshaping it. With Korea now in the mix, the East Asian open-source ecosystem is forming a triangular dynamic: China, Korea, and (let's not forget) the rest of the world watching closely.

What does this mean for developers and businesses? For one, more choices. Open-source models like K-EXAONE 2.0 offer an alternative to proprietary systems, and with Apache 2.0 licensing, there are few restrictions on usage. That's a win for innovation.

But it also raises the bar. As models grow larger and more capable, the competition intensifies, pushing everyone to improve. For users, that translates to better AI tools, faster.

The Road Ahead

LG's move is a calculated one. By open-sourcing K-EXAONE 2.0, they're not just sharing technology—they're building an ecosystem. Developers can now build on this foundation, creating applications that could rival anything from Silicon Valley or Shenzhen.

Of course, size isn't everything. The real test will be how well K-EXAONE 2.0 performs in real-world scenarios, beyond benchmarks. But early signs are promising.

As the AI race accelerates, one thing is clear: the era of open-source AI is no longer dominated by a few players. With Korea stepping up, the future looks more diverse—and more exciting.

Key Points

  • K-EXAONE 2.0 is South Korea's largest AI model, with 750 billion parameters (37B active).
  • Fully open-sourced under Apache 2.0, a rare move for Korean AI.
  • Performance gains: +10% average benchmark score, +30% in coding/agent tasks.
  • Competitive edge: Outperforms GLM-5.1 in long context, and both GLM-5.1 and Qwen3.5 in agent tool use.
  • Strategic significance: Signals Korea's entry into the global open-source AI race, forming a new East Asian triad.