On November 9, 2026, LG AI Research announced at NVIDIA AI Day Seoul 2026, held at COEX in Seoul, that the API for its foundation model K-EXAONE 2.5 is now free. Developers, companies, and research institutions can call the 750-billion-parameter model for service development and research at no cost, simply by obtaining an API key on the EXAONE API Platform. It is the first time LG has handed its flagship-class model to outside developers as a free API — an invitation to try it first and talk business later.
What exactly is free
K-EXAONE 2.5 is a technology-validation model built by further training the second-generation model LG released in August, not a brand-new generation. According to LG's model card, it has 750 billion total parameters with 31 billion active per token, a 512K-token context window, and a knowledge cutoff of January 2026. It handles text only, in Korean and English, and supports chat, tool calling, structured outputs, code generation, and streaming. The endpoint is OpenAI-compatible under the model ID k-exaone_v2.5, so existing clients can switch over by changing the address. LG says overall performance across 24 evaluation metrics improved by more than 70% in eight weeks, and it published four agentic benchmark scores: 74.6 on AutomationBench-Openset, 59.2 on SWE-Bench Pro, 78.0 on BrowseComp, and 77.3 on Claw-Eval — claiming performance comparable to frontier open-weight models such as DeepSeek-V4.1-Flash and GLM-5.3. These are LG's own internal evaluations, however, and have not yet been independently verified.
Free does not mean without limits
For developers, free means the cost of experimentation drops to zero: no business negotiation, no waiting for a weights download — just test the agentic and coding abilities of a 750-billion-parameter model directly. That suits Korean teams and smaller companies that want to validate an idea fast. The limits are clear, though: free is not open source, and the 2.5 weights have not been released. It supports only Korean and English text, so it is no help for Chinese-language use cases. At the same trillion-parameter scale, Mistral Large 4 went the open-weights route, while LG is opening only a free API and keeping the weights in its own hands. LG has also not said how long the free period will last, or what pricing and rate limits will apply once it ends — something to plan around before running production workloads on it.
LG says it is now focused on developing the third-generation model, aiming to complete a frontier-class EXAONE model by January 2027, and plans to use NVIDIA's Dynamo technology to improve inference efficiency and expand service capacity. The free release of 2.5 looks like a public rehearsal: before the frontier model arrives, LG wants enough developers putting its validation technology through its paces.
