LG Opens K-EXAONE 2.5 API Free After 70% Performance Jump in 8 Weeks
LG AI Research said it has raised the performance of its proprietary AI foundation model by more than 70% in eight weeks and is opening the upgraded “K-EXAONE 2.5” to anyone through a free API, Maeil Business Newspaper (MT) reported.
The announcement came on the 9th at “NVIDIA AI Day Seoul 2026,” held at COEX in Seoul. Developers, startups, companies and research institutions can obtain an API key on LG AI Research’s platform and use K-EXAONE 2.5 at no cost for service development and research.
Benchmarks Place K-EXAONE 2.5 Among Frontier Open-Weight Models
K-EXAONE 2.5 is a technical validation model built on the second proprietary foundation model LG AI Research released in August, with additional training applied. On open datasets, it posted results competitive with frontier-level global open-weight AI models in agentic AI, the source reported.
LG AI Research published scores on four agentic AI benchmarks, comparing K-EXAONE 2.5 with China’s DeepSeek V4.1 Flash and GLM 5.3 (Max):
LG Opens Free K-EXAONE 2.5 API as It Explores Nvidia Dynamo LG AI Research has announced free API access to K-EXAONE 2.5, giving developers, startups, businesses, and researchers a new way to experiment with its AI model. The immediate benefit is a lower initial cost barrier to… pic.twitter.com/RE8jnq1DGZ
| Benchmark | K-EXAONE 2.5 | DeepSeek V4.1 Flash | GLM 5.3 (Max) |
|---|---|---|---|
| Automation Bench (SaaS task automation) | 74.6 | 74.9 | 71.6 |
| Claw Eval (workplace problem solving) | 77.3 | 75.8 | 77.6 |
| BrowseComp (web research) | 78.0 | 75.8 | 79.0 |
| SWE-Bench Pro (software engineering) | 59.2 | 58.4 | 64.0 |
Kim I-reun, head of LG AI Research’s EXAONE Lab, said, “Internal analysis confirmed that performance across 24 evaluation metrics improved by more than 70% compared to the model released in August.”
NVIDIA Dynamo Collaboration Under Review
LG AI Research is also reviewing the use of NVIDIA’s “Dynamo,” a high-performance operating system, to raise the inference efficiency and service scalability of its AI model. NVIDIA has already published a Dynamo usage guide tailored to K-EXAONE, the report said.
Third Model Targeted for January
The company plans to fold the data quality, post-training and reinforcement-learning techniques validated with K-EXAONE 2.5 into the development of its third model.
Im Woo-hyung, co-research director at LG AI Research, said, “We are concentrating our capabilities on developing the third model and expect this API release to let more developers and companies directly experience the changed performance.” He added that the goal is to complete an EXAONE model with frontier-level performance by January next year.
The push follows LG Group Chairman Koo Kwang-mo’s New Year address this year, in which he named “selection and concentration” as the start of innovation and stressed that once a choice is made, it must be pursued “to a level others consider impossible.”
K-EXAONE 2.5 Free API: What to Know
What is K-EXAONE 2.5? It is a technical validation model from LG AI Research, built on the foundation model released in August with additional training, and designed to let AI agents plan, use tools and complete real-world tasks.
How do you access the free API? Anyone can request an API key through LG AI Research’s own platform starting the 9th and use the model at no cost, according to the report.
How does it compare with rivals? On the four published benchmarks, it scored 74.6 on Automation Bench, 77.3 on Claw Eval, 78.0 on BrowseComp and 59.2 on SWE-Bench Pro — within range of DeepSeek V4.1 Flash and GLM 5.3 (Max) on each test.
What comes next? LG AI Research said it will apply the validated training techniques to a third model and aims to finish a frontier-level EXAONE model by January next year.
Keep reading