EXAONE API Platform

EXAONE API Platform This is a model API platform provided by LG AI Research for developers, currently featuring the K-EXAONE 2.5 large language model and the EXAONE Tabular predictive model. On November 9, 2026, the platform announced the launch of free APIs, suitable for non-commercial research and testing of Korean and English text, code, and agent capabilities.Official Release Notes

Please review the scope of authorization before use:The current Terms of Service are limited to non-commercial use; they prohibit the use of the model or its output for commercial purposes, as well as the use of products or services that allow third parties to call the model or submit input. Teams planning to integrate the model into paid applications, customer service, or publicly available tools must first obtain the appropriate authorization from the platform; they cannot assume that commercial use is permitted solely based on the “free API” designation.Article 2 of the Terms of Service

Product and Team Background

The platform is provided by LG AI Research, and API documentation, model descriptions, and integration guides are available at docs.exaone.ai. K-EXAONE 2.5 is a proof-of-concept model that has been further trained based on previous versions. In this release, the official team is highlighting its capabilities in reasoning, programming, and the use of agent tools. The API platform serves as the entry point for cloud-based calls and is not equivalent to model weights for download or software packages that can be freely redistributed.LG AI Research Release Notes

core functionality

• Text Comprehension and Generation:K-EXAONE 2.5 primarily supports Korean and English and can be used for question-and-answer tasks, text organization, and content analysis. When Chinese language capabilities are required, you should first evaluate the model using your own training data; do not directly apply the official descriptions of its Korean and English capabilities to Chinese.
• Code and Agent Tasks:It provides capabilities for code generation, analysis, debugging, tool invocation, and structured output, making it suitable for researching how models break down multi-step tasks into executable operations. Whether the tool actually executes and which systems it can access still depends on the implementation of the integration endpoint.
• Long-form text processing:Supports 512K contexts. The interface specifies that the sum of the input token count and `max_tokens` must not exceed 524,288; neither the input nor the output may be calculated as a full 512K at the same time.
• Streaming Response and Inference Modes:It can return results in standard JSON or as a stream; the API offers "low," "high," and "max" inference intensities. The model always performs inference and cannot be switched to a non-inference model by turning off "thinking mode."
• Standalone Table Model:EXAONE Tabular is designed for structured data forecasting. Tabular forecasting and the text interface of K-EXAONE 2.5 are distinct functionalities; users should select the appropriate model based on the task at hand.

For model capabilities, seeModel Overview; For context and reasoning parameters, seeK-EXAONE 2.5 Interface Documentation.

For Which Situations and Audiences Is It Suitable?

• Researchers and Trainees:Use publicly available, synthetic, or licensed data to conduct research on long-text processing, Korean-English question-answering, and code generation.
• Agent Framework Developers:Validate OpenAI-compatible APIs, inference content processing, tool call chains, and exception retries; do not expose unauthorized services to third parties.
• Technical teams focusing on Korean language models:Include it in the evaluation tests, compare its accuracy, processing time, and output stability on our own tasks, and then decide whether to discuss commercial use.

If your requirements involve directly deploying commercial applications, stable large-scale production calls, or image or audio input, you should prioritize verifying authorization and product capabilities. The current API for K-EXAONE 2.5 is a text-only interface.

How to Get Started

1. show (a ticket)EXAONE API Platform, log in to your account, read the Terms of Service, and request an API key.
2. Store the secret in a backend environment variable or a secret management system; do not place it in the web front end, mobile app packages, or public code repositories. The API uses a Bearer token for authorization.Authentication Instructions
3. Follow the official example to send a request to https://api.exaone.ai/v1/chat/completions, specifying "k-exaone_v2.5" as the model name. For your first test, start with short text and a small output budget to verify the returned results and error handling.
4. If you already have OpenAI-compatible call code, adjust the service URL, model name, and parameters according to the API documentation. The platform also provides integration guides for Hermes and OpenCode, which are suitable for further testing of agent workflows.Agent Onboarding Guide
5. Read the remaining quota and retry instructions from the response; when encountering a 429 error, wait according to the rate-limiting policy, and do not attempt to circumvent the limit by using multiple accounts or multiple keys. Periodically check the key expiration date.

Fees, Limits, and Usage Restrictions

To the extent permitted by law, users retain the right to input data and own the generated output, while granting the platform a perpetual, worldwide, non-exclusive, royalty-free license to use such output. Consent for training may be withdrawn under “My Profile > Consent and Notification,” or by submitting a request to privacy@lgresearch.ai. Upon withdrawal, the input and output will no longer be used for training; however, portions already incorporated into the model’s training cannot be separated or deleted from the model. For details, seeArticle 2 of the Terms of Servicecap (a poem)Section 6 of the Privacy Policy.

See the section on credit limits and options for details.Speed Limit Document, see "Use of Data" for detailsprivacy policy; Account and eligibility are subject toTerms of Service (Korean)The English version shall prevail.

Comparison with Similar Model APIs

When making a selection, first determine whether the license covers your intended use, and then compare task quality, API compatibility, latency, and total cost. EXAONE’s free tier is ideal for lowering the barrier to entry for research, but it cannot replace the licensing and stability arrangements required for commercial deployment.