LG EXAONE
LG's open-weight models, from a 33B vision model to a 750B Apache-licensed flagship
Overview
EXAONE is LG AI Research's family of language and vision-language models, first unveiled in December 2021. The current releases are EXAONE 4.5, a 33B open-weight vision-language model under a non-commercial licence, and K-EXAONE 2.0, a 750B mixture-of-experts model released in July 2026 under Apache 2.0. Weights are downloaded from Hugging Face; hosted inference runs on FriendliAI dedicated GPU endpoints.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- 750B parameters, 37B active
- 262,144-token context
- Ten languages, Korean to Vietnamese
- 33B vision-language model with FP8, AWQ and GGUF builds
- EXAONE 4.0 1.2B on-device model
- Runs in vLLM, SGLang, TensorRT-LLM and llama.cpp
- Single-tenant GPU inference
- Container option for your own environment
- Required for any revenue-generating use of EXAONE 4.5 or earlier
Use cases
What you can produce with LG EXAONE
- K-EXAONE 2.0 weights (750B MoE, Apache 2.0) for commercial self-hosting
- EXAONE 4.5 33B vision-language weights for document, chart and image understanding in research
- A 262,144-token context on both current models, with reasoning and non-reasoning modes
- Ten-language text generation in K-EXAONE 2.0, including Japanese and Vietnamese
- A compact EXAONE 4.0 1.2B model for on-device use
- OpenAI-compatible serving through vLLM, SGLang or TensorRT-LLM
- Dedicated hosted endpoints through FriendliAI, billed by GPU-hour
ASEAN Perspective
LG EXAONE in Southeast Asia
EXAONE 4.0 does not currently support any Southeast Asian language — Vietnamese, Thai, Indonesian, Malay, Filipino, and other regional languages are absent — making it a weaker out-of-the-box choice for ASEAN-language NLP tasks compared to Alibaba Qwen or SeaLLMs. That said, its strong math, science, and coding benchmarks are language-agnostic, and ASEAN enterprises running Korean-joint-venture or Korean-tech supply-chain operations may find the model's Korean-English bilingual strength directly useful. LG has a substantial hardware and electronics presence across Southeast Asia (consumer electronics, home appliances, automotive), and as the EXAONE ecosystem matures — particularly the on-device 1.2B variant for embedded applications — ASEAN device manufacturers partnering with LG could encounter EXAONE in embedded product contexts before any cloud deployment arrives.
K-EXAONE 2.0 is the headline: a 750B mixture-of-experts model under Apache 2.0, so it can be used commercially without asking, with ten languages and strong long-context scores in LG's own tests. EXAONE 4.5 is the more practical size, a 33B vision-language model that fits on one H200 and reads documents and charts well by LG's account, but its non-commercial licence rules out production use without a deal.
Tooling is thinner than for Qwen or Llama, there is no per-token API since FriendliAI dropped EXAONE from its serverless list, and Vietnamese is the only ASEAN language. A serious research option, and for K-EXAONE 2.0 a genuinely open one if you have the GPUs.
What we found
EXAONE is the model family of LG AI Research, the AI institute LG set up in December 2020. It was unveiled at the LG AI Talk Concert on 14 December 2021 as a multimodal "super-giant" model and has been rebuilt several times since: EXAONE 3.0 in 2024 was the first version with downloadable weights, followed by 3.5, the EXAONE Deep reasoning models in March 2025 and the hybrid-reasoning EXAONE 4.0 in July 2025, in 32B and 1.2B on-device sizes.
Two releases define it in 2026. EXAONE 4.5, published in April, is LG's first open-weight vision-language model: a 31.7B language model plus a 1.29B vision encoder, with a 262,144-token context and six languages, Vietnamese among them. LG says it beats GPT-5 mini, Claude Sonnet 4.5 and Qwen3-VL on its average across 13 vision benchmarks; those are LG's own evaluations. Its weights stay under the EXAONE AI Model License 1.2-NC, which allows research and education only, so commercial use needs a separate agreement with the licensor, LG Management Development Institute.
K-EXAONE 2.0, released on 31 July under the government's Sovereign AI Foundation Model Project, went the other way. It is a 750B-parameter mixture-of-experts model with 37B active, which LG calls the largest foundation model built in Korea, it covers ten languages, and it ships under Apache 2.0, so commercial use is unrestricted. LG reports a 70.1 average across 24 benchmarks, up from 63.3 for the first K-EXAONE. In August K-EXAONE passed the project's second-stage evaluation, and in September LG showed EXAONE-based expert models for tabular data, materials discovery and finance.
The practical catches: K-EXAONE 2.0 needs a multi-GPU cluster and, at release, forks of vLLM and SGLang; LG's public K-EXAONE 2.0 demo was offline when checked in October 2026; and FriendliAI, the API partner LG links, now offers EXAONE only on dedicated GPUs billed by the hour rather than per token. Vietnamese remains the only ASEAN language.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Notable facts
- EXAONE 4.0's 32B model scored 94.5 on South Korea's university entrance exam (CSAT) math section and 90 on the 2024 US Mathematical Olympiad.
- The name EXAONE is an acronym for 'EXpert AI for EveryONE', reflecting LG's goal of broad enterprise deployment rather than a single consumer product.
- EXAONE 4.0's 1.2B on-device model is half the size of its predecessor (EXAONE 3.5 2.4B) yet outperforms OpenAI's GPT-4o mini on specialist math and coding evaluations.
- LG AI Research's founding president Kyunghoon Bae was nominated as South Korea's Minister of Science and ICT in 2025, underscoring the model's national-strategic significance.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including LG EXAONE's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with LG EXAONE unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to LG EXAONE directly →
Spotted something out of date? Suggest an update →
More in LLMs & Chat
