LG AI Research opened API access to K-EXAONE 2.5 and EXAONE Tabular on November 9, 2026, according to its platform changelog. The launch gives developers a hosted route to language-model and structured-data experiments, subject to registration, usage limits and noncommercial terms.
Two models become available through the EXAONE API
The changelog lists K-EXAONE 2.5 as a 750-billion-parameter language model and EXAONE Tabular as a model for tabular prediction. LG’s model catalogue lists Korean and English support for K-EXAONE 2.5, alongside chat, tool calling, structured outputs, code generation and streaming.
The two interfaces serve different research tasks. K-EXAONE 2.5 accepts conversational text; EXAONE Tabular takes files containing labelled examples and rows for prediction. Their documented inputs and outputs make that distinction concrete:
| Model | Documented input | Documented output |
|---|---|---|
| K-EXAONE 2.5 | Text messages sent to the chat-completions endpoint. | Generated text, with reasoning returned separately from the final answer. |
| EXAONE Tabular | Training and test CSV files, with a target column specified. | Classification labels or class probabilities; regression point predictions or a predictive distribution. |
The language-model reference describes an OpenAI-compatible API and a context limit of 524,288 tokens, shared between input and the requested maximum generation budget. Reasoning counts toward generated tokens. The Tabular reference documents separate classification and regression endpoints. LG says this model supports prediction without dataset-specific model training.
Access comes with noncommercial and integration limits
Developers use an API key as a bearer token, as shown in LG’s Quick Start. The platform terms, effective November 9, grant noncommercial use and prohibit using models or outputs commercially. They also prohibit enabling third parties to call a model or submit inputs, including through products or services integrated with the API.
Those conditions matter for universities considering a shared campus application or an externally accessible research service. The opening does not itself establish permission for those deployments. The terms also say that the platform supplies no model weights, parameters or source code, and that API keys expire 30 days after issuance.
LG’s rate-limit documentation describes request windows measured by second, minute and day. Excess requests return HTTP 429, while response headers report remaining allowances. These limits make available throughput a separate consideration from the model’s context capacity.
What the launch means for research evaluation
For researchers assembling a model evaluation suite, the launch offers two distinct kinds of experiment: language and coding tasks through chat completions, and structured-data prediction through CSV uploads. The documentation identifies the supported interfaces; their usefulness for a particular study still depends on results from that study’s own tasks.
The separation of reasoning from final answers has a practical implication for a reproducible AI experiment: an evaluation record needs to distinguish which output it scores. LG also exposes reasoning intensity as a request setting. Together, these details make request configuration part of interpreting any reported result.
For research offices, data reuse is another material part of the launch. LG says consent to usage-data analysis for model training and service improvement is optional, increases the daily allowance, and is unnecessary for continued API access. That connects submission choices to training-data provenance. Separately, section 7.8 of the platform terms prohibits including another person’s personal information in inputs, a direct constraint on the research material that can be submitted.
Sources
- LG AI Research: EXAONE API Platform changelog, November 9, 2026. Checked 2026-11-14.
- LG AI Research: Models. Checked 2026-11-14.
- LG AI Research: K-EXAONE 2.5 API reference. Checked 2026-11-14.
- LG AI Research: EXAONE Tabular API reference. Checked 2026-11-14.
- LG AI Research: Quick Start. Checked 2026-11-14.
- LG: EXAONE API Platform Terms of Use, effective November 9, 2026. Checked 2026-11-14.
- LG AI Research: Rate Limits. Checked 2026-11-14.
