Motif Technologies Raises Bridge Round for Motif 4, Nvidia to Join

Motif 4 targets a top-five global ranking

Motif 4 succeeds Motif 3, which was unveiled in August. It is scheduled for release as an open-weight model in the first quarter of next year. Open-weight means the trained model's weights (parameters) are published so anyone can download, use and modify them. The company says it aims to place Motif 4 among the world's top five models.

Motif 3 was built under the Ministry of Science and ICT's "independent AI foundation model" program. According to the company, it scored 47 on the composite intelligence index of evaluator Artificial Analysis in August, ranking 10th among all AI models and 4th among open-weight models. That was the highest placing outside models from US and Chinese companies. Motif says it used self-designed architecture, including Grouped Differential Latent Attention (GDLA), to build Motif 3 in five months on 768 GPUs.

Fixing the post-training gap

Limited time and resources meant Motif 3 did not get enough post-training, such as reinforcement learning. Post-training is the stage that refines a pretrained model for practical use. With the new funding, the company plans to strengthen post-training for Motif 4 to improve real-world usability and commercial appeal.

CEO Jung-hwan Lim, who holds a mathematics doctorate from Oxford, said that in a fiercely competitive global model race, a company must differentiate at every stage, from architecture to post-training, to survive. He said Motif 3 fell short on post-training because of deadline and resource limits, and promised that Motif 4 would deliver not only technical excellence but also usability and commercial viability.