other
7.8B parameters
Restricted
Reviewed May 2026

EXAONE Deep 7.8B

EXAONE Deep 7.8B is LG AI Research's reasoning-focused model, fine-tuned from EXAONE-3.5-7.8B-Instruct for math and coding tasks. It claims benchmark wins over OpenAI o1-mini on AIME 2024 and MATH-500 at this parameter count. Context window is 32,768 tokens; quantized GGUF and AWQ builds are available.

License: exaone·Context: 32,768 tokens
BLK · VERDICT

Our verdict

OP · Fredoline Eruo|VERIFIED MAY 28, 2026
9.3/10

If you need a sub-10B reasoning model for math or code and you're working in Korean or English, EXAONE Deep 7.8B is a legitimate option — the o1-mini benchmark comparisons are credible and the quantized formats keep VRAM requirements manageable. The hard blocker is the license: non-commercial only, so any revenue-generating use requires going back to LG AI Research. For personal projects or research, it's worth a look; for anything commercial, skip it until the licensing situation changes.

Why this rating

Auto-generated rating (Opus 4.7 judge, claude-opus-4-7). Overall 9.25/10. License claim matches the HF card (license: other, license_name: exaone) and the non-commercial restriction is correctly flagged with explicit caution in the verdict. Metadata is verifiable directly from the card: 7.8B params, 32,768 context, base model relation, Korean/English. Editorial voice is honest and operator-grade — names the o1-mini benchmark claim as a claim, flags trust_remote_code as a security consideration, and gives a clear commercial-use blocker. Best use case is sharp (Korean/English math and coding reasoning), and weaknesses are concrete. Brand fit is solid: a popular sub-10B reasoning model with available GGUF/AWQ quantizations is squarely in runlocalai's wheelhouse.

Overview

EXAONE Deep 7.8B is LG AI Research's reasoning-focused model, fine-tuned from EXAONE-3.5-7.8B-Instruct for math and coding tasks. It claims benchmark wins over OpenAI o1-mini on AIME 2024 and MATH-500 at this parameter count. Context window is 32,768 tokens; quantized GGUF and AWQ builds are available.

Strengths

  • Outperforms o1-mini on AIME 2024 and MATH-500 benchmarks at 7.8B params
  • Reasoning-tuned — better suited to multi-step math and code than a vanilla instruct model
  • 32K context handles lengthy chain-of-thought without truncation
  • AWQ and GGUF quantizations available for lower-VRAM deployments

Weaknesses

  • Non-commercial EXAONE license — not usable in production without explicit LG AI Research approval
  • Requires trust_remote_code=True, which is a security consideration in shared or multi-tenant environments
  • Trained on Korean and English only; other languages will degrade noticeably
  • 167K HF downloads is modest — community troubleshooting resources are limited

Quantization variants

Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.

QuantizationFile sizeVRAM required
Q4_K_M4.3 GB6 GB

Get the model

HuggingFace

Original weights

huggingface.co/LGAI-EXAONE/EXAONE-Deep-7.8B

Source repository — direct quantization required.

Hardware that runs this

Cards with enough VRAM for at least one quantization of EXAONE Deep 7.8B.

Compare alternatives

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 Deep 7.8B?

6GB of VRAM is enough to run EXAONE Deep 7.8B at the Q4_K_M quantization (file size 4.3 GB). Higher-quality quantizations need more.

Can I use EXAONE Deep 7.8B commercially?

EXAONE Deep 7.8B is released under the exaone, which has restrictions for commercial use. Review the license terms before using it in a product.

What's the context length of EXAONE Deep 7.8B?

EXAONE Deep 7.8B supports a context window of 32,768 tokens (about 33K).

Source: huggingface.co/LGAI-EXAONE/EXAONE-Deep-7.8B

Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.

Related — keep moving

Before you buy

Verify EXAONE Deep 7.8B runs on your specific hardware before committing money.