LG AI Research has officially released the API for its proprietary foundation model, K-EXAONE 2.5, an enhanced iteration of the version previously unveiled in August. This release aims to foster innovation by providing developers, startups, enterprises, and research institutions with free access to the model via the EXAONE API Platform.

Core Focus: Agentic AI Performance

K-EXAONE 2.5 is specifically engineered to function as an Agentic AI, exhibiting advanced capabilities in autonomous reasoning, task automation, and complex problem-solving. Performance evaluations based on open datasets indicate that K-EXAONE 2.5 achieves parity with competitive frontier-level global open-weight models, specifically demonstrating strong performance against models such as DeepSeek-V4.1-Flash and GLM-5.3.

Technical Benchmarking

The model’s capabilities were rigorously validated across four key benchmarks, quantifying its proficiency in specialized agentic tasks:

• AutomationBench (Openset): Achieved a score of 74.6, measuring the model’s ability to perform task automation using SaaS (Software as a Service) tools.
• SWE-Bench Pro: Achieved a score of 59.2, evaluating the model’s capacity to execute real-world software development, including code modification and architectural adjustments.
• BrowseComp: Achieved a score of 78.0, assessing the efficiency and accuracy of the model in navigating web environments to retrieve specific, actionable information.
• Claw-Eval: Achieved a score of 77.3, which evaluates the model’s general problem-solving proficiency within multifaceted professional work environments.

Practical Applications and Software Productivity

Beyond theoretical benchmarking, K-EXAONE 2.5 is designed to enhance the software development lifecycle (SDLC). The model provides specialized utility in:

1. Code Generation and Debugging: Facilitating the rapid construction of functional code blocks and identifying logical errors.
2. Refactoring and Maintenance: Improving existing codebases through systematic optimization.
3. Test Code Generation: Automating the creation of test suites to ensure application reliability.

By integrating these capabilities, K-EXAONE 2.5 enables developers to abstract away complexity, thereby accelerating the transition from conceptual ideation to full-scale application deployment. Access is currently available to the public via official API key issuance, positioning the model as a versatile tool for both academic research and commercial software automation.