What happened
LG AI Research made its K‑Exaone 2.5 publicly accessible through a free API starting on Nov. 9, 2026, at Nvidia AI Day Seoul. The move is part of South Korea’s “independent AI foundation model” program and is intended to collect real‑world usage data to improve the next‑generation model, K‑Exaone 3.0. LG also disclosed a parallel technology cooperation with Nvidia, evaluating Nvidia’s Dynamo infrastructure to boost inference efficiency and scalability. The company released its own benchmark results, showing K‑Exaone 2.5 scoring competitively on several agentic‑AI tests such as Automation Bench (74.6), ClawEval (77.3), BrowseComp (78.0) and SWE‑Bench Pro (59.2). Access to the API is open to developers, startups, enterprises and research institutions via an API key obtained from the Exaone website.
At Nvidia AI Day Seoul on Nov. 9, LG AI Research announced that the K‑Exaone 2.5 model—an iteration of its independent AI series—would be made available to anyone via a free API. Interested parties can register on the Exaone website to receive an API key, enabling them to call the model for service development or research purposes.
The release is tied to South Korea’s national program to develop independent AI foundation models. LG’s K‑Exaone 2.5 is described as a “technology‑validation” model that builds on the second independent model unveiled in August, focusing on refining training methodology and identifying performance gaps before the next major version, K‑Exaone 3.0.
LG provided its own benchmark results: Automation Bench (74.6), ClawEval (77.3), BrowseComp (78.0) and SWE‑Bench Pro (59.2). These scores place K‑Exaone 2.5 on par with or slightly ahead of Chinese models DeepSeek V4.1 Flash and GLM 5.3 on most metrics, indicating competitive capability in agentic‑AI tasks such as autonomous planning, web browsing, and software code modification.
In parallel, LG disclosed a technology cooperation with Nvidia. The company is evaluating Nvidia’s Dynamo infrastructure, which distributes large‑scale inference workloads to reduce cost and improve scalability. If adopted, Dynamo could help LG keep the free API financially viable as usage grows.
LG executives emphasized that the free API serves a dual purpose: providing developers and researchers immediate access to a high‑performing model, and generating real‑world usage data that will inform the training pipeline for K‑Exaone 3.0, which LG aims to launch with “frontier‑level” performance by January 2027.
Source details: finance.biggo.com
Why it matters
The free release of a domestically built marks a significant shift in South Korea’s AI strategy, offering an alternative to costly foreign APIs and potentially lowering entry barriers for local innovators. By opening K‑Exaone 2.5 without a paywall, LG aims to generate large‑scale, real‑world interaction data that can be fed back into the training pipeline of K‑Exaone 3.0, accelerating its path to “frontier‑level” performance. The benchmark scores suggest the model can compete with leading Chinese offerings such as DeepSeek V4.1 Flash and GLM 5.3, positioning South Korea as a contender in the rapidly evolving “agentic AI” space where models plan and execute tasks autonomously. The partnership with Nvidia on Dynamo technology could also demonstrate a path to cost‑effective, high‑throughput inference, addressing a common hurdle for free‑tier AI services. However, the sustainability of a free API depends on how LG manages inference costs and whether the model can maintain performance parity as larger, better‑funded competitors continue to iterate.
The launch directly addresses a common barrier for Korean startups and academic groups: the high cost of accessing leading AI models from the United States or China. By offering a no‑cost alternative, LG potentially accelerates local AI product development and research, fostering a more self‑sufficient ecosystem.
Benchmark performance suggests K‑Exaone 2.5 can hold its own against established Chinese models, which is notable given the rapid pace of AI advancement globally. This could shift competitive dynamics in the Asian AI market, where Chinese open‑source models have previously dominated the cost‑sensitive segment.
Collecting large‑scale interaction data is a proven method for improving foundation models. The free API creates a feedback loop that LG can exploit to refine training data pipelines, reinforcement‑learning strategies, and overall model before the commercial launch of K‑Exaone 3.0.
The Nvidia Dynamo partnership signals a strategic alignment with leading hardware and inference‑optimization technology. If successful, it may set a precedent for other Korean AI firms to leverage similar infrastructure, reducing the financial burden of offering free services at scale.
Policy implications are also significant. The South Korean government’s push for independent AI models aims to reduce reliance on foreign AI providers for national security and economic reasons. LG’s public API demonstrates tangible progress toward that policy goal, potentially influencing future funding allocations and regulatory frameworks.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Which component of an AI application is the machine-learning model itself?
What to watch next
Key indicators to monitor include the volume and diversity of API usage, any announced pricing or tiered‑service changes, and the technical outcomes of the Nvidia Dynamo integration. Follow LG’s updates on K‑Exaone 3.0, slated for a January 2027 release, to see whether the data collected from the free API translates into measurable performance gains. Additionally, watch for policy responses from South Korea’s Ministry of Science and ICT, which may adjust funding or regulatory support based on the model’s adoption and competitive standing against foreign AI ecosystems.
API adoption metrics: number of registered developers, request volume, and diversity of use cases will indicate market impact and data collection success.
Any shift in access policy: while the API is free now, LG may introduce tiered pricing or usage limits if inference costs become unsustainable.
Technical outcomes of the Dynamo integration: performance benchmarks, cost per inference, and scalability reports will reveal whether the partnership delivers the promised efficiency gains.
Timeline and feature set of K‑Exaone 3.0: announcements about its performance, availability, and pricing will show how effectively LG leverages the data gathered from the 2.5 version.
Government response: updates from the Ministry of Science and ICT regarding funding, regulatory support, or additional national AI initiatives could amplify or constrain LG’s strategy.