AI model development startup Trillionlabs has decided to actively participate in an additional call for proposals for the Ministry of Science and ICT’s “independent AI foundation model” programme, which the ministry is expected to announce soon.

Trillionlabs said it has developed foundation models using a from-scratch approach, independently carrying out the full process from model design to training, rather than reworking existing open-source models or fine-tuning external models.

Since its founding, it has maintained that development approach and built in-house technologies for its own architecture design, data collection and refinement, and its pre-training and post-training pipelines.

Trillionlabs stressed that, unlike many companies that claim to develop proprietary models but disclose only limited information about development processes, datasets and training methods, it has continuously shared technical assets gained during development with the open-source community.

It said it opens not only weights but also algorithms and checkpoints.

Weights are numeric values a model obtains through learning and uses to process input data and predict outputs. A checkpoint is a file that saves a model’s state at a specific point in time, including weights, and can be used to resume training after an interruption or for deployment. The company stressed that 공개ing checkpoints would also be a major help to AI research.

Trillionlabs said it plans to use this achievement to strengthen transparency in developing independent foundation models and expand a global verification structure centered on open source. It said it aims to go beyond simply following global standards and to continuously spread the authenticity and reliability of South Korean AI technology on the international stage through an open ecosystem that developers worldwide can directly verify.

According to statistics from the “Korean AI open-source heatmap” published on the global open-source sharing platform Hugging Face, Trillionlabs recorded 22 technical contributions over the past year, ranking third among domestic companies.

It was the third-highest figure, following LG AI Research with 33 contributions and second-ranked Naver with 27. Jae-min Shin (신재민), chief executive of Trillionlabs, said, “It is meaningful that Trillionlabs’ technology has been verified on a global platform as a true independent model born from unique design capabilities, not simple fine-tuning of external models.” He added, “Having proven our technical capabilities through Hugging Face, which developers around the world trust, we will continue to contribute to the global open-source ecosystem beyond Korea.”