The exam ecosystem is developing with new AI models

July 15, LG Ai Research–The R&D AI arm from South Korea LG group–Unveiled Exaone 4.0A model of hybrid IA reasoning which combines general language treatment with advanced reasoning capacities introduced through previous society Deep model exam.

LG AI Research claims that its new model surpasses the similar models of Alibaba, Microsoft and Mistral AI in industry benchmarks for science, mathematics and coding. However, Exaone 4.0 is still not part of the best Deepseek model.

However, LG AI research does not pursue the same users as most familiar names in AI. Unlike models such as Chatgpt and Gemini, which are mainly designed for the average person, LG I target professional users. “Our main objective is the business-business sector (B2B) rather than consumer business [for now]”Explains Honglak Lee, the new co-chief of LG Ai Research and former researcher at Google Brain. LG launched the company in December 2020 as part of the digital transformation strategy of the Korean technology giant.

To this end, LG AI Research has made examination 4.0 available for research and academic use on hugs, the global open source IA platform. The model also supports the use of the Spanish language, expanding its capacities beyond its original skills with Korean and English.

Exona and strategic roadmap ecosystem

Only a week after the start of Exaone 4.0, research on LG AI made its commitment to a B2B clear orientation, revealing its wider exam ecosystem and its strategic roadmap. At the Talk 2025 on July 22, the company revealed several new models.

Among the models are Exaone 4.0 Vision Language, a multimodal AI model that can interpret both text and images, and examination Path 2.0, a model focused on health care designed to diagnose patient conditions in minutes. There are also several company -specific AI agents: Chatexaone, an agent currently used internally by LG employees to support business workflows; Foundry of ExaONE data, a platform to accelerate the generation of data; And a complete on -site agent which can be deployed in isolated secure environments without exposing sensitive data.

LG AI Research indicates that Exaone 4.0 VL, which will be launched in the near future, ends with Meta’s Llama 4 Scout in performance tests. In addition, the company claims that Data Foundry can do in a single day which generally takes 60 experts three months.

The on -site on -site agent works on fleas developed by Furiosaai, a startup startup based in South Korea Neuronal treatment units (NPU) adapted to the workloads of the AI. According to the company, the Furiosaai RNGD accelerator provided inference performance on the Exaone models 2.25 times faster than the competitors.

LG also says that the equipment is more energy efficient. A single rack powered by RNGD chips can generate up to 3.75 times more tokens for exams an exams than a traditional gpu rack operating within the same power limits.

Autonomous agents for business security

The ultimate objective of LG AI Research is to provide companies with all the main components necessary to securely execute autonomous agents in their own infrastructure, with an integrated generation of data and commercial operating features, lee on ieee Spectrum told.

“We don’t only offer an inference engine,” says Lee. “We aim to provide an end -to -end system that really integrates the key features that companies really need – so that they can immediately connect it to their workflow. Each company has unique operational needs. This is why we design our solution to be flexible – to combine and configure different parts depending on the environment of each customer. ”

Further on, the company lays the foundations of physical AI or AI integrated into robots. “Physical AI is still in its infancy,” explains Lee. “But the central framework – perception, reasoning and action in a continuous loop – is something that we actively build.”

Although the company does not yet apply this directly to robots, it demonstrates the same loop with Chatexaone, or Agent Nexus, an AI agent designed to assess the legal compliance of data sets. The Nexus is crucial for the ability to crawl on the Internet. “These agents must understand the web pages, extract relevant information and act on them,” explains Lee. “This is why we build web agents who can navigate in complex information flows and make independent decisions.”