LG opens free API access to K-EXAONE 2.5, its 750-billion-parameter Korean AI model

LG AI Research announced on Monday, November 9, at NVIDIA AI Day Seoul 2026 at COEX in Seoul that it is opening free API access to K-EXAONE 2.5, its flagship language model. Developers, startups, companies and research institutions can request an API key through the EXAONE API Platform and use the model for service development and research at no charge, starting today.

What is being opened for free

The release covers hosted API access to K-EXAONE 2.5, not open model weights. The platform offers an OpenAI-compatible chat completions endpoint under the model identifier k-exaone_v2.5, so existing client code can point at it directly. LG also opened EXAONE Tabular, a small model for predictions on structured data. LG has not said how long the free period will last, nor what usage limits will apply once it ends.

750 billion parameters, 512K context

According to LG's model card, K-EXAONE 2.5 has 750 billion parameters with 31 billion active, handles text only in Korean and English, and offers a 512,000-token context window (524,288 tokens for input plus maximum generation), with a knowledge cutoff of January 2026.

LG describes version 2.5 as a technology-validation model: further post-training applied to K-EXAONE 2.0, the second-generation foundation model released in August, rather than a wholly new generation. LG AI Research says performance improved by more than 70 percent against the August model across a combined assessment of 24 evaluation metrics.

The 70 percent gain is an internal figure

That improvement is LG's own internal measurement, with the baseline and evaluation details unpublished and no independent verification yet. LG says the point of the free release is to let more people try the model in real services and build up validation data ahead of a frontier-level EXAONE model it aims to complete by January next year. LG also says it is considering NVIDIA's Dynamo technology to improve inference efficiency and service scalability, a possible direction rather than a confirmed deployment.

The same event also saw another Korean sovereign-AI announcement: NVIDIA, Rebellions and SK Telecom said they would jointly build sovereign AI infrastructure in Korea, which this site reported earlier today.