Atlas

Models

← All models

EXAONE 4.0 1.2B

LG AI Research · 2025-07-15 · 1.3B parameters

EXAONE 4.0 1.2B is LG AI Research's small dense causal language model for on-device applications, released with the 32B EXAONE 4.0 model. It supports English, Korean, and Spanish, combines switchable non-reasoning and reasoning modes with agentic tool use, and provides a 65,536-token context after progressive long-context extension.

Benchmark scores

BenchmarkScore
AA-LCR0.0
AA-Omniscience Index-82.4
AIME 202550.3
Artificial Analysis Intelligence Index2.9
Artificial Analysis Omniscience Accuracy6.2
Artificial Analysis Omniscience Hallucination Rate91.5
Artificial Analysis Openness Index27.8
Capability127.1
CritPt0.0
GPQA Diamond51.5
Humanity's Last Exam (Text-Only)5.8
IFBench (Artificial Analysis)25.3
LiveCodeBench (Artificial Analysis)51.6
SciCode (Artificial Analysis)9.3
Terminal-Bench Hard0.0
τ²-bench (Telecom)20.5
Loading Atlas data…