Leaderboard / LG AI Research (EXAONE)

EXAONE 4.0 1.2B

Name: EXAONE 4.0 1.2B
Author: LG AI Research (EXAONE)

LG AI Research (EXAONE) · EXAONE 4.0 · released 2025-07-15 · data checked 2026-11-01

EXAONE 4.0 1.2B is a 1.3B-parameter dense model from LG AI Research (EXAONE) (Rest of Asia), released 2025-07-15 under the EXAONE License (non-commercial) license. It scores 36.1 on the Small LM Index, #13 of 21 tiny models. It needs about 1.8 GB of memory at 4-bit with an 8K-token context.

Small LM Index

36.1

#13 of 21 scored tiny models

Likely 32.7–39.6 · from 4 of 6 core benchmarks

Parameters

1.3B

Dense

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

64K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best tiny model on that benchmark. Sources: 9 reported by LG AI Research (EXAONE). How we label sources.

MMLU-Pro Lab-reported Knowledge59.3

Class best: Qwen3.5 4B 79.1

GPQA Diamond Lab-reported Knowledge52.0

Class best: Nanbeige4.2 3B 87.4

AIME Lab-reported Math · AIME 202545.2

Class best: Spark-X2.6 4B 91.2

LiveCodeBench Lab-reported Coding · LiveCodeBench v645.3

Class best: LFM3 1.7B 73.5

MMLU Lab-reported Knowledge · MMLU-Redux71.5

Class best: MiniCPM5 2B 84.7

Multilingual MMLU Lab-reported Multilingual · MMMLU Spanish only62.4

Class best: SmolLM3 3B 64.1

IFEval Lab-reported Instruction following67.8

Class best: LFM3 1.7B 96.7

BFCL Lab-reported Agentic · BFCL v352.9

Class best: Spark-X2.6 4B 71.8

HMMT Lab-reported Math · HMMT Feb 202534.0

Class best: Nanbeige4.2 3B 82.8

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

Humanity's Last Exam≈ 3

SWE-bench Verified≈ 17

Specifications

Lab: LG AI Research (EXAONE) (KR · Rest of Asia)
Architecture: Dense transformer
Total parameters: 1.3B
Active parameters: All (dense)
Size class: Tiny
Context window: 64K tokens
Input modalities: text
Reasoning: Hybrid (thinking on/off)
License: EXAONE License (non-commercial)
Openness: Open weights
Memory to run: ~1.8 GB at 4-bit, 8K context (more sizes)
Hugging Face: 13.3K downloads (30 days) · 193 likes

Weights on Hugging Face ↗

Compare with

EXAONE 4.0 1.2B vs LFM3 1.7B (1.7B, index 60.0)
EXAONE 4.0 1.2B vs Spark-X2.6 1.7B (1.7B, index 50.1)
EXAONE 4.0 1.2B vs LFM2.5 1.2B Thinking (1.2B, index 30.1)
EXAONE 4.0 1.2B vs LFM2.5 1.2B Instruct (1.2B, index 20.6)
EXAONE 4.0 1.2B vs Gemma 3 1B (1B, index 6.1)
EXAONE 4.0 1.2B vs Qwen3.5 0.8B (900M, index 10.7)

Sources

https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-1.2B

Something wrong here? Report it.

Run it locally

Uses LGAI-EXAONE/EXAONE-4.0-1.2B-GGUF (Q4_K_M). Swap the tag for another size: BF16, IQ4_XS, Q4_K_M, Q5_K_M, Q6_K, Q8_0.

Ollama

ollama run hf.co/LGAI-EXAONE/EXAONE-4.0-1.2B-GGUF:Q4_K_M

llama.cpp

llama-server -hf LGAI-EXAONE/EXAONE-4.0-1.2B-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model mlx-community/exaone-4.0-1.2b-4bit

vLLM (GPU server)

vllm serve LGAI-EXAONE/EXAONE-4.0-1.2B

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	64K context
16-bit F16	3.3 GB	5.1 GB	7.0 GB
8-bit Q8_0	2.1 GB	3.9 GB	5.8 GB
4-bit Q4_K_M	1.6 GB	3.3 GB	5.2 GB

Check it against your GPU, Mac or phone →