K-EXAONE 2.5 API Is Free: How to Try LG’s Model

this article contains no affiliate links. All links are plain links, and we earn nothing if you click them. How this article was prepared: AI helped gather and draft this research; a person checked the facts against the cited sources.

Quick answer: On November 9, 2026, LG AI Research opened the K-EXAONE 2.5 API at no charge to developers, startups, companies and researchers. Grab a key at docs.exaone.ai and hit an OpenAI-compatible endpoint with the model name k-exaone_v2.5. It’s text-only, capped at 524,288 tokens, and rate-limited. Read the official Terms of Service before you put it in anything public or paid.

What LG announced on November 9

LG AI Research opened the K-EXAONE 2.5 API for free on November 9, 2026.

K-EXAONE 2.5 is a “technical validation model.” LG says it aims to finish a frontier-level EXAONE model by January 2027.

How to try it

We did not walk through the sign-up screens. These steps follow only the official docs and the LG release.

1. Get an API key. Open the EXAONE docs, sign in, and create a key. Every request needs that key as a Bearer token in the Authorization header. Keep it off public pages and out of browser code.

2. Call the chat endpoint. Send a POST request to https://api.exaone.ai/v1/chat/completions and set model to k-exaone_v2.5. The docs say the API is OpenAI-compatible. The model is text-only.

3. Watch the rate-limit headers. Going over your limit returns a 429 error. Responses carry rate-limit headers for the limit and what you have left. Opting in to usage-data consent raises the daily API usage limit.

![Official LG AI Research docs page for K-EXAONE 2.5 showing the heading, a POST /v1/chat/completions badge, model ID k-exaone_v2.5 in the Request table, and the 524,288 token limit note](https://techkit.cloud/wp-content/uploads/2026/11/k-exaone-2-5-docs.png)

Source: docs.exaone.ai — K-EXAONE 2.5, captured Nov 10, 2026. Credit: LG AI Research. Cropped official docs page. Not a runtime test.

Limits worth knowing

Max context is 524,288 tokens — input plus max_tokens. Go over that and you get a 400. The model always reasons before it answers; thinking cannot be turned off. Set reasoning_effort to low, high or max (default max). Reasoning lands in reasoning_content.

LG has not published an end date for free access. The Terms say fees will be posted if the company starts charging.

Read the terms first

We opened the official EXAONE API Platform Terms of Service. Section 2.1 grants a non-exclusive, non-transferable, revocable right to use the platform for non-commercial purposes. It also says you may not use any AI model or its output for commercial purposes, or let a third party call a model or send it input, including by integrating the API into your products or services.

That sits next to LG’s press release, which says anyone with a key can use K-EXAONE 2.5 for service development and research at no extra cost. We are not guessing how those two statements fit together. Read the Terms yourself before you build anything with the API.

Two other rules on the same page: API keys last 30 days from issuance (section 3.4), and you must be 19 or older (section 8.1).

How it compares (LG’s own tests)

The scores below come from LG AI Research. We have not run independent tests. LG compared K-EXAONE 2.5 with DeepSeek V4.1 Flash and GLM 5.3 (max):

| Benchmark | K-EXAONE 2.5 | DeepSeek V4.1 Flash | GLM 5.3 (max) | | — | — | — | — | | AutomationBench | 74.6 | 74.9 | 71.6 | | Claw-Eval | 77.3 | 75.8 | 77.6 | | BrowseComp | 78.0 | 75.8 | 79.0 | | SWE-Bench Pro | 59.2 | 58.4 | 64.0 |

!TechKit text card of LG's own benchmark scores for K-EXAONE 2.5 versus DeepSeek V4.1 Flash and GLM 5.3 (max) on AutomationBench, Claw-Eval, BrowseComp and SWE-Bench Pro, labelled LG internal evaluation

TechKit text card. Numbers from LG AI Research, checked Nov 17, 2026. Labelled LG internal evaluation — we did not run these tests.

LG also says overall performance is more than 70% better than the August model. Treat that as a company claim.

Quick questions

Is K-EXAONE 2.5 really free?

LG says the API is free as of November 9, 2026, and has not published an end date. The Terms say fees will be posted if charging begins.

How do I get an API key?

Through the official docs at docs.exaone.ai. Keys expire after 30 days.

Can I use it in a commercial product?

The official Terms say the license is for non-commercial use and that you may not use the model or its output commercially, or let a third party call the API through your product. LG’s announcement, separately, talks about free use for service development and research. Read the Terms before you decide.

Is it better than DeepSeek?

On LG’s own four tests it sits close to DeepSeek V4.1 Flash, and behind GLM 5.3 (max) on two of them. That is LG’s evaluation, not an independent review.

Sources

LG press release: K-EXAONE 2.5 API opened for free (Korean, 2026.11.09) — https://www.lg.co.kr/media/release/30760

K-EXAONE 2.5 API docs — https://docs.exaone.ai/en/docs/k-exaone-2.5

EXAONE rate limits — https://docs.exaone.ai/en/docs/rate-limits

EXAONE authentication — https://docs.exaone.ai/en/docs/authentication

EXAONE API Platform Terms of Service — https://docs.exaone.ai/en/terms/terms-of-service

Last checked: November 17, 2026.

Get new free-deal guides & reviews by email

1 thought on “K-EXAONE 2.5 API Is Free: How to Try LG’s Model”

Leave a Comment