Skip to content
← Models
Open weightsLG AI Research/10 trustMetrics-checked

EXAONE 3.5

South Korea's open-source flagship trio — 2.4B on-device, 7.8B general, 32B high-performance.

Visit official page →

About

EXAONE 3.5, open-sourced December 9, 2024, is LG AI Research's flagship multimodal model release and a point of national pride in Korea's sovereign-AI push. Unlike its predecessor (which open-sourced only a lightweight variant), LG released all three sizes: a 2.4B ultra-lightweight model for on-device use, a 7.8B general-purpose model, and a 32B high-performance model.

LG's technical report claims top scores among global open-source models across 20 benchmarks covering real-world usability, long-text processing, coding, and math. Two design emphases stand out: Retrieval-Augmented Generation over real-time web results or uploaded documents to reduce hallucination, and multi-step reasoning that decomposes queries methodically. Long-context handling reaches roughly 100 pages per session, with a planned expansion to 128K tokens in 2025. The open release is research-oriented; commercial deployments require engaging LG directly.

Specs

Context window

Pricing

Free quota

API access

Weights

Open weights

Released

December 9, 2024

Benchmarks

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

korean-nlpragon-deviceenterprise

Trust score — 7/10

/10 trustMetrics-checked

Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 59; Trending score 72/100 from researcher assessment.

More in the exaone family

Sources