November 9, 2026 — LG AI Research has opened free API access to K-EXAONE 2.5, the latest version of its sovereign foundation model, through its EXAONE API Platform. Anyone — developers, startups, companies and research institutions — can get an API key and start using the model at no cost. Why it matters: a 750-billion-parameter model that LG says now trades blows with leading Chinese open-weight systems on agentic benchmarks is suddenly free to test in your own products.
What happened
LG announced the move on November 9 at NVIDIA AI Day Seoul 2026, held at COEX in Seoul. K-EXAONE 2.5 is what LG calls a technology-validation model: it applies additional post-training to K-EXAONE 2.0, the second-generation model released in August 2026 under South Korea’s sovereign AI foundation model programme. The point of this release is real-world validation — LG wants outside developers putting the model to work so the techniques tested here can feed into its next-generation model.
The details
According to LG’s model documentation, K-EXAONE 2.5 has 750 billion total parameters with 31 billion active per token. It is text-only, works in Korean and English, and offers a 512K-token context window, with a January 2026 knowledge cutoff. The API is OpenAI-compatible — a standard POST /v1/chat/completions endpoint using the model identifier k-exaone_v2.5 — and supports chat, tool calling, structured outputs, code generation and streaming, so existing OpenAI-based tooling can point at it with minimal changes. The platform also lists EXAONE Tabular, a smaller model for predictions on structured data.
On performance, LG says its internal analysis across 24 evaluation metrics shows an improvement of more than 70% over the model released in August. In the agentic benchmarks LG published, K-EXAONE 2.5 scores 59.21 on SWE-Bench Pro (agentic coding, up from 47.1 for version 2.0), 78.0 on BrowseComp (agentic search), 74.6 on AutomationBench and 77.29 on Claw-Eval — figures that put it roughly level with DeepSeek’s V4.1 Flash and Zhipu’s GLM-5.3 on several of those tests, ahead on some and behind on others. Those are LG’s own measurements on open datasets, not independent results, and they should be read that way. LG has not said how long the free period will last. Separately, LG said it is reviewing NVIDIA’s Dynamo technology to improve the model’s inference efficiency — a possible future step, not a confirmed deployment.
Why it matters
Sovereign AI models usually stay behind government and enterprise doors. LG is doing the opposite: handing a national-champion model to any developer with an API key, for free, in a format that drops into the tooling most teams already use. If the benchmark claims hold up in production, this is a genuinely competitive free option for agentic coding and workplace automation — and a useful pressure test for the paid APIs. LG AI Research head Im Woo-hyung said the lab is focused on its third-generation model and aims to complete a frontier-level EXAONE model by January 2027; the free 2.5 API is how it gathers real-world evidence on the way there. Our take, in the spirit of everything AI, tested: free and OpenAI-compatible means there is no excuse not to try it against your current model on your own tasks — just don’t retire the paid API on the strength of vendor benchmarks alone.
Frequently asked questions
Is K-EXAONE 2.5 really free to use?
Yes, for now. LG says developers, startups, companies and research institutions can obtain an API key through the EXAONE API Platform and use the model for service development and research at no cost. LG has not announced how long the free period lasts or what limits may apply afterward.
How do I access the K-EXAONE 2.5 API?
Sign up for an API key on LG’s EXAONE API Platform (docs.exaone.ai) and call its OpenAI-compatible chat completions endpoint with the model name k-exaone_v2.5.
What can K-EXAONE 2.5 do?
It is a text-only model for Korean and English with a 512K-token context window. LG highlights code generation, debugging and refactoring, agentic search and workplace automation, with support for tool calling, structured outputs and streaming.
Are the benchmark results independently verified?
No. The scores — including the claimed improvement of more than 70% over the August model across 24 metrics — come from LG’s own internal measurements. Independent testing will show how the model performs outside LG’s lab.
Sources: LG AI Research announcement via Digital Today, Money Today and Aju Press (November 9, 2026); EXAONE API Platform documentation; Unite.AI; Spectrum Web Co.