LG AI Research has opened free API access to its K-Exaone 2.5 foundation model, giving developers and research groups a new South Korean model to test without an API bill. The company says the release also helps validate training methods intended for its next model, K-Exaone 3.0.
Watch Desk analysis
What happened
LG AI Research announced the API opening on 9 November at Nvidia AI Day Seoul. According to BigGo Finance, users can obtain an API key through the Exaone website for service development and research. The company describes K-Exaone 2.5 as a technology-validation model, developed through additional training on the second model in its series.
LG AI Research’s own evaluations report scores of 74.6 on Automation Bench, 77.3 on ClawEval, 78.0 on BrowseComp and 59.2 on SWE-Bench Pro. The report says those results compare favourably with named competitor models on several benchmarks. They are company-reported scores, not independent tests. LG says performance across 24 benchmarks improved by more than 70% compared with its model released in August.
The company plans to carry data-design and reinforcement-learning techniques tested with K-Exaone 2.5 into K-Exaone 3.0. It is also reviewing Nvidia Dynamo for inference efficiency and service scalability; that adoption is under consideration, not confirmed.
Why it matters
Free API access lowers the cost of trying a model for startups, researchers and developers who might otherwise stick with established providers. It also gives LG a route to gather real-world usage experience as it develops its next model. The practical test will be whether the model is useful in actual applications, not merely competitive on a benchmark table.
Our read
This is a useful invitation to experiment, with a clear caveat: the performance figures come from LG’s own evaluations. Developers can try the API now, but should compare it on their own tasks before treating benchmark rankings as a buying guide. Free access is a welcome door; it is not, by itself, a quality certificate.
What to watch
- How access, usage limits and any later pricing are set out for API users.
- Whether independent evaluations support LG’s reported benchmark results.
- What training methods and performance LG carries into K-Exaone 3.0.
Discussion spark: Would free API access persuade you to test a newer model, or do independent evaluations need to come first?
Sources and evidence
- South Korea's LG AI Research Opens 'K-Exaone 2.5' API Free to All, Partners with Nvidia – BigGo Finance (9 November 2026, 01:25 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.