LG reveals in South Korea a new model for Amnesty International, which is Exaone 4.0, describing it as the first hybrid AI system in the country.
On Tuesday, the company announced that New model It can produce and evaluate the hypotheses.
In her press statement, LG also confirmed that she will continue to work on improving Exaone.lee Jin Sik, head of Exaone Laboratory at LG AI Research, “We will continue our efforts to research and develop to establish Exaone as a major border model for Korea and show its competitiveness in the global market.”
LG AI Research Exaone launched at the depths of March
LG AI Research was first established in 2020 to accelerate the practical use of Amnesty International. It was not long ago, the group launched its first model, Exaone 1.0, in 2021.
After that, in March 2025, the company launched the first artificial intelligence model in South Korea (AI), Exaone Deep, with its main model, Exaone Deep-32B.
The model includes 32 billion teachers to help learn artificial intelligence and thinking, and experts claim that it leads at a similar level to Deepseek’s R1. The Ri’s Deepseek model has 671 billion teachers, or about 95 % more than Exaone Deep-32-B.
While the highest number of parameters is often associated with improving performance, it also requires more arithmetic resources. As a result, more companies convert their focus towards improving models to provide similar results with fewer parameters.
In many performance tests, the Exaone Deep R1, DeepSeek’s R1 and alibaba’s exaltation model exceeds Qwq-32B models Most. In the year 2024 of the American Mathematics Olympiad, Exaone Deep-32B 90, overlooking Deepseek and alibaba at 86.7.
He also led the South Korean model in the South Korean mathematics section 2025 with impressive 94.5. Not to mention this, it outperformed the QWQ-32B version in the science level, and got a degree of 66.1 compared to 63.3. However, Exaone Deep is still lagging behind in claims and language compared to Alibaba models in Deepseek.
According to LG, the new Exaone 4.0 performance was better in more performance tests than other artificial intelligence models from the United States, China and France. To date, very few companies have developed similar mixed models. The established hybrid thinking systems include Claude models and Alibaba’s QWEN. Aside from it, the new Exaone 4.0 will be available in Seoul on July 22.
Openai has released two artificial intelligence models
Openai has also launched new models of artificial intelligence, O3 and O4-MINI. The company explained that the O3 model has been improved to solve complex scientific problems, although it trades the depth speed in the treatment. Codex Cli also includes an artificial intelligence agent designed to help people with coding. The open source feature can work directly on the user’s computer, and work through the peripheral device program.
Meanwhile, the O4-MINI model performs similarly to O3, although it is more compact and smart. Both models accept visual representation in the thinking process, which is the first for any Openai thinking system. According to the company, the systems can enlarge or process images to connect accurate responses. To date, both models can be reached for the company’s subscribers.
The CEO of the company, Sam Altman, is also preparing to issue an open version of the male thinking system in the coming months, and see Deepseek’s success through the open source model, R1. Altman has also stated that they would launch the GPT-5 within a few months.
The main difference wireCrypto Secret Tool projects are used to get guaranteed media coverage
