LG AI Research has unveiled EXAONE 4.5, a multimodal AI model that combines text and image understanding to outperform larger competitors from OpenAI, Google, and Alibaba across 13 visual assessment benchmarks. The model scored 77.3 on five key STEM (Science, Technology, Engineering, and Mathematics) benchmarks, surpassing OpenAI's GPT-5-mini (73.5), Anthropic's Claude 4.5 Sonnet (74.6), and Alibaba's Qwen-3-VL (77.0) . Released on April 9, 2026, EXAONE 4.5 represents a significant step forward in how AI systems process and reason about visual information in professional settings.

What Makes EXAONE 4.5 Different From Other Multimodal AI Models?

EXAONE 4.5 is built as a Vision-Language Model (VLM), which means it integrates a proprietary vision encoder with a Large Language Model (LLM) into a single unified architecture. In simpler terms, the model can look at an image and understand what it sees, then reason about that information just as well as it processes written text. This dual capability is crucial for real-world business applications where documents often mix text, charts, diagrams, and photographs .

What sets EXAONE 4.5 apart is its efficiency. Despite having only 33 billion parameters, roughly one-seventh the size of LG's larger "K-EXAONE" model released in late 2025, EXAONE 4.5 achieves comparable performance in text comprehension and reasoning. This breakthrough stems from LG AI Research's proprietary Hybrid Attention architecture and high-speed inference technology based on multi-token prediction, which allows the model to process information faster without sacrificing accuracy .

Where Does EXAONE 4.5 Excel in Real-World Tasks?

The model demonstrates exceptional performance on tasks that matter to enterprises. On LiveCodeBench v6, a benchmark for coding ability, EXAONE 4.5 scored 81.4, exceeding Google's Gemma 4 (80.0). More impressively, on ChartQA Pro, which tests the ability to analyze and reason through complex charts, the model recorded 62.2, the highest score among models in its class .

The real strength lies in document comprehension. EXAONE 4.5 excels at accurately reading and reasoning through complex documents encountered in real-world industrial settings, such as contracts, technical drawings, financial statements, and scanned documents. This capability addresses a genuine pain point for enterprises that rely on automated document processing .

How to Deploy EXAONE 4.5 in Your Organization

• Access the Model: EXAONE 4.5 is available on Hugging Face, a global open-source platform, permitting use for research, academic, and educational purposes without licensing fees.
• Evaluate Document Processing Needs: Assess whether your organization handles contracts, technical drawings, financial statements, or scanned documents that could benefit from automated visual reasoning and comprehension.
• Test on Benchmark Tasks: Run pilot projects on coding tasks, chart analysis, or document interpretation to determine whether EXAONE 4.5's performance gains justify integration into your existing AI infrastructure.
• Consider Language Support: EXAONE 4.5 supports Korean, English, Spanish, German, Japanese, and Vietnamese, making it viable for multilingual enterprises operating across these regions.

LG AI Research has also expanded language support beyond Korean and English to include Spanish, German, Japanese, and Vietnamese, broadening the model's applicability for global enterprises .

What Does This Mean for the Future of AI in Business?

EXAONE 4.5 represents LG AI's entry into what the company calls the "multimodal era." The model is positioned as a stepping stone toward a broader vision: evolving EXAONE into a form of "Physical Intelligence," an AI capable of understanding and making judgments within the physical world, transcending the boundaries of virtual environments . This roadmap suggests that future versions will incorporate audio, video, and environmental understanding, enabling AI to make practical judgments and take action within industrial settings.

"EXAONE 4.5 represents LG AI's successful entry into the multimodal era, where AI understands not just text, but visual information as well. Starting with this model, we will expand AI's scope of understanding to include audio, video, and the physical environment, ultimately creating AI that can make practical judgments and take action within industrial settings," explained Jinsik Lee, Head of EXAONE Lab at LG AI Research.

Jinsik Lee, Head of EXAONE Lab at LG AI Research

Beyond raw performance, LG is also investing in cultural and contextual understanding. The company trained EXAONE using data provided by the Northeast Asian History Foundation and is discussing collaborations with other domestic institutions holding high-quality data. This effort reflects a broader recognition that language models need to understand cultural nuance, not just linguistic patterns .

"The surge in AI models capable of speaking Korean does not equate to a true understanding of cultural sensitivity. With its built-in K-AUT (Korea-Augmented Universal Taxonomy), EXAONE is evolving to provide expressive depth alongside robust reliability, setting a new standard for culturally-aware AI," noted Myoungshin Kim, Head of the AI Safety and Trust Office at LG AI Research.

Myoungshin Kim, Head of the AI Safety and Trust Office at LG AI Research

LG is also nurturing the next generation of AI developers. The company hosted the "LG Aimers" Hackathon, a program dedicated to nurturing young AI experts, with a focus on developing lightweight versions of the EXAONE model. By releasing EXAONE 4.5 as an open-weight model on Hugging Face, LG is following through on its commitment to expand the AI research ecosystem, a strategy it began with EXAONE 3.0 in August 2024, which was Korea's first open-weight multimodal AI model .

The competitive landscape for multimodal AI is intensifying. EXAONE 4.5's performance gains over larger models from OpenAI, Google, and Alibaba suggest that efficiency and specialized optimization may matter more than raw parameter count. For enterprises evaluating multimodal AI solutions, this development signals that smaller, well-engineered models can deliver competitive results while consuming fewer computational resources, reducing both deployment costs and environmental impact.