K-EXAONE 2.5 at a glance
LG AI Research made the EXAONE API Platform generally available on November 9, 2026, free of charge. Its flagship K-EXAONE 2.5 is a 750B-parameter (31B active) Korean-English agentic model on an OpenAI-compatible API, launched alongside EXAONE Tabular.
What changed with K-EXAONE 2.5
LG AI Research opened its EXAONE foundation models to any developer on November 9, 2026, through the EXAONE API Platform, and the platform changelog states the service is currently available free of charge. The flagship is K-EXAONE 2.5, a text-only mixture-of-experts model with 750 billion total and 31 billion active parameters, built for Korean and English agentic work: coding, tool calling, structured outputs and long documents. Its OpenAI-compatible endpoint (model ID k-exaone_v2.5) allows a maximum context of 524,288 tokens, its knowledge cutoff is January 2026, and a reasoning_effort setting offers low, high and max, with max as the default. The launch also includes EXAONE Tabular, a prediction model for structured data. Korean press reported the announcement was made at NVIDIA AI Day Seoul 2026.
In LG AI Research's own benchmark table, K-EXAONE 2.5 scores 59.21 on SWE-Bench Pro (DeepSeek-V4.1-Flash 58.41, GLM-5.3 64.03), 78.0 on BrowseComp, 74.6 on AutomationBench Open-set and 77.29 on Claw-Eval: broadly level with the open-weight models LG compares it against, ahead on some and behind on others, and a large jump over K-EXAONE 2.0. These are the vendor's numbers, not independent tests: as of November 9, 2026, Artificial Analysis had not published a K-EXAONE 2.5 page. LG has not announced when the free service ends or what paid tiers will follow, and API rate limits apply.
K-EXAONE 2.5 in numbers
Values are copied from the linked source on the date shown. They are not IA Compa measurements.
- Context windowas of · Source for Context window: docs.exaone.ai (opens in a new tab)
- 524,288 tokens
- SWE-Bench Pro (LG AI Research's own evaluation)as of · Source for SWE-Bench Pro (LG AI Research's own evaluation): docs.exaone.ai (opens in a new tab)
- 59.21
- BrowseComp (LG AI Research's own evaluation)as of · Source for BrowseComp (LG AI Research's own evaluation): docs.exaone.ai (opens in a new tab)
- 78.0
- AutomationBench Open-set (LG AI Research's own evaluation)as of · Source for AutomationBench Open-set (LG AI Research's own evaluation): docs.exaone.ai (opens in a new tab)
- 74.6
- Claw-Eval (LG AI Research's own evaluation)as of · Source for Claw-Eval (LG AI Research's own evaluation): docs.exaone.ai (opens in a new tab)
- 77.29
How it compares
Current models we have covered: released in the last six months, newest version of each line only. Each value is the model's latest recorded value. (Models)
Where to compare K-EXAONE 2.5
Independent leaderboards and official resources for comparing this model yourself. Rankings change often; each site shows its own current data.
General leaderboards
Why K-EXAONE 2.5 matters for businesses
Why it matters: a frontier-scale agentic model at zero API cost changes the math for prototyping. Agencies can test agent workflows, Korean-English products and sovereign-AI pitches without a token bill, then decide whether the platform earns a paid place. The cautions are real — vendor-only benchmarks, a new platform and no stated end date for the free tier — but as a free option next to DeepSeek and GLM, it is worth a test project.
Get AI News by email
Verified AI stories, each with its official source, on Monday and Thursday mornings. Add breaking news to hear about the biggest ones as soon as we publish them.
Every story links its official source
Free, in English or Spanish
Unsubscribe with one click, any time
Sources for this story
- EXAONE API Platform changelog, November 9, 2026 (opens in a new tab)Official
LG AI Research · docs.exaone.ai
- K-EXAONE 2.5 model page, EXAONE API Platform (opens in a new tab)Official
LG AI Research · docs.exaone.ai
- K-EXAONE 2.5 endpoint reference, EXAONE API Platform (opens in a new tab)Official
LG AI Research · docs.exaone.ai
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #free-api
- #mixture-of-experts
- #agentic-ai
- #sovereign-ai
- #korean-ai
- #openai-compatible-api
