What changed
LG AI Research made K-EXAONE 2.5 available through the EXAONE API Platform at no charge with an API key. The model is an additional-training technical-validation version based on its August K-EXAONE release and is positioned for agent and coding workflows.
Why it matters
The API lowers the cost of testing a large Korean-English model without operating the earlier 750B-parameter weights. It is useful as an evaluation endpoint, but its validation label and unknown long-term limits make it a poor default dependency without an exit path.
What to do
Obtain a key, pin the exposed model identity and run Korean, English, tool-use and coding holdouts with latency, rate-limit, privacy and output-quality logs; keep a portable fallback until production terms and limits are explicit.
Caveat
LG's agent benchmark scores are internal runs on public datasets and have not been independently reproduced. The announcement does not establish durable free pricing, service-level commitments, retention terms or an open K-EXAONE 2.5 checkpoint.