Skip to content

An editorial team and a community building with AI

vibecoding.incJoin Discord ↗
← All articles

News · LG AI Research · 2 min read

Event date: · Published:

LG AI Research opens free API access to K-EXAONE 2.5

LG AI Research released free API access to K-EXAONE 2.5, a post-trained update to its national AI model, claiming agentic-coding benchmark scores competitive with DeepSeek-V4.1-Flash and GLM-5.3.

On November 9, 2026, LG AI Research opened free API access to K-EXAONE 2.5, a post-trained update to its national-project foundation model, according to LG AI Research's newsroom announcement. The release was unveiled at NVIDIA AI Day Seoul 2026, held at COEX. K-EXAONE is Korea's national AI foundation model project, backed by the Ministry of Science and ICT (MSIT).

What LG AI Research changed

  • K-EXAONE 2.5 is described as a technical validation model, built through additional post-training on the K-EXAONE 2.0 base released July 31, 2026.
  • Free access is available immediately through the EXAONE API Platform at docs.exaone.ai, by registering for an API key.
  • LG reports a combined performance improvement of more than 70% across 24 evaluation metrics, compared with the August model.
  • Benchmark scores given: AutomationBench-Openset 74.6, SWE-Bench Pro 59.2, BrowseComp 78.0, and Claw-Eval 77.3.
  • The underlying K-EXAONE 2.0 base model is a 750-billion-parameter hybrid-attention mixture-of-experts model with 37 billion active parameters.

What it means for AI builders

LG AI Research names developers, startups, enterprises, and research institutions as the model's target users. The company highlights capabilities including complex reasoning, code generation, debugging, refactoring, and test creation, plus web navigation, information retrieval, and SaaS task automation. LG positions K-EXAONE 2.5 as competitive with the Chinese open models DeepSeek-V4.1-Flash and GLM-5.3 on agentic tasks. That gives builders a free, benchmarked agentic-coding model from a major lab outside the US and China, widening the pool of frontier-adjacent models for building coding agents, and signals Korea's state-backed push for sovereign AI competitive with Chinese open models.

Availability and next steps

LG AI Research plans to fold the validated training techniques from K-EXAONE 2.5 into a frontier-class K-EXAONE 3.0 model, targeted for completion by January 2027.

LG has secured core training technology in-house to respond to a rapidly changing global benchmark environment, said EXAONE Lab head Kim Yi-reun.

Sources

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.

LG AI Research opens free API access to K-EXAONE 2.5 | vibecoding.inc