EXAONE 3.5 32B Instruct
EXAONE 3.5 32B Instruct is LG AI Research's 32B bilingual model, trained for instruction-following in English and Korean. It supports a 32,768-token context window. The license is non-commercial only, so business use is off the table.
If you're doing non-commercial Korean-English NLP work and have the VRAM for a 32B model, EXAONE 3.5 is worth a look — it's one of the few serious Korean-bilingual options at this scale. That said, the non-commercial-only license is a hard blocker for anything production-facing, so don't build on it if there's any chance your use case turns commercial. The vendor's own warnings about bias and factual errors are standard boilerplate, but worth taking seriously at this size. Hedge: useful for researchers, a skip for anyone shipping a product.
›Why this rating
Auto-generated rating (Opus 4.7 judge, claude-opus-4-7). Overall 9.00/10. The row is largely accurate and well-voiced: license is correctly flagged as non-commercial (EXAONE AI Model License), params/context match the card, and the verdict is honest about the production blocker. However, practical deployability falls short — there's no mention of VRAM requirements for a 32B model (≈64GB+ bf16, or quantized alternatives), no note on trust_remote_code requirement, and no mention of GGUF availability. The bestUseCase is reasonable but slightly broad ('research and personal projects'); a sharper domain anchor would help. Family is set to 'other' which is defensible since EXAONE is its own architecture, but worth confirming. Just below the 9.0 bar — fix deployability detail and it passes.
Flags: - No VRAM/quantization guidance for a 32B model — critical for local-AI readers - Missing mention of trust_remote_code=True requirement - bestUseCase could be sharper than 'research and personal projects' - No mention of whether GGUF/quantized community ports exist
Overview
EXAONE 3.5 32B Instruct is LG AI Research's 32B bilingual model, trained for instruction-following in English and Korean. It supports a 32,768-token context window. The license is non-commercial only, so business use is off the table.
Strengths
- Bilingual English and Korean instruction-following
- 32,768-token context window handles long documents
- 32B parameter scale puts it in the serious general-purpose tier
- 11,500+ HF downloads suggests active community testing
Weaknesses
- Non-commercial license — cannot be used in production products or services
- Known risk of biased or factually incorrect outputs, per vendor's own documentation
- Knowledge cutoff means it will be wrong or stale on recent events
- 129 HF likes is modest for a 32B model, indicating limited community validation so far
Quantization variants
Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.
| Quantization | File size | VRAM required |
|---|---|---|
| Q4_K_M | 17.6 GB | 23 GB |
Get the model
HuggingFace
Original weights
Source repository — direct quantization required.
Hardware that runs this
Cards with enough VRAM for at least one quantization of EXAONE 3.5 32B Instruct.
Models worth comparing
Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.
Frequently asked
What's the minimum VRAM to run EXAONE 3.5 32B Instruct?
Can I use EXAONE 3.5 32B Instruct commercially?
What's the context length of EXAONE 3.5 32B Instruct?
Source: huggingface.co/LGAI-EXAONE/EXAONE-3.5-32B-Instruct
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.
Related — keep moving
Verify EXAONE 3.5 32B Instruct runs on your specific hardware before committing money.