EXAONE Deep 32B
Korean-strong reasoning model in 32B, 7.8B, and 2.4B sizes
About
EXAONE Deep is LG AI Research's reasoning model family, released at NVIDIA GTC 2025 in 32B, 7.8B, and 2.4B sizes. It is built on LG's EXAONE foundation with particular strength in Korean, and the company reports strong math results — 95.7 on MATH-500 and top marks on Korea's 2025 CSAT math sections. The smaller sizes are the practical story: the 7.8B keeps up to 95% of the 32B's performance at a quarter of the size, per LG. The weights are open but under LG's custom EXAONE license, not an OSI-approved license, so commercial users need to read the terms. English-language documentation and community tooling are thinner than for the Qwen or Llama ecosystems.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
LG-reported: MATH-500 95.7, GPQA Diamond 66.1, LiveCodeBench 59.5, AIME 2024 90.0
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: LG AI Research model card and arXiv paper (2503.12524) read: 32B/7.8B/2.4B sizes, released at GTC 2025 / March 18 2025.; Benchmarks LG-reported: MATH-500 95.7, GPQA Diamond 66.1, LiveCodeBench 59.5, AIME 2024 90.0, CSAT 94.5.; License caveat: custom EXAONE license, not OSI-approved — flagged in description.; Limitation: thinner English docs and community tooling than Qwen/Llama ecosystems..