South Korea's Ministry of Science and ICT on Thursday announced the results of an additional call for its Indigenous AI Foundation Model project. Motif Technologies was selected as the final winner, and the project will enter a second round competition among 4 elite teams.

The ministry held a briefing at the Government Complex Seoul on Thursday. It said it selected Motif Technologies as an additional elite team after a written review and presentation assessment.

The project will now move into its second round with 4 teams: LG AI Research, Upstage, SK Telecom and Motif Technologies. The existing 3 teams will develop AI models from January to June, while Motif Technologies will do so from February to July. All 4 teams will undergo a second-stage evaluation around early August.

Motif Technologies plans to start with a 300B-class reasoning large language model to build an AI foundation model aimed at strengthening national competitiveness. It plans to progressively advance to a 310B-class vision language model and a 320B-class vision language action model. The consortium includes 17 organisations, including Moreh, Crowdworks, Seoul National University, KAIST, Hanyang University, Kyunghyang Shinmun, and Samil PwC.

Evaluators assessed Motif Technologies as having a high level of technological internalisation. They said it had experience proposing and implementing many core modules in-house, and had developed proprietary foundation models not only for text but also for images and video.

Trillion Labs, which competed alongside it, was not selected by a narrow points margin. The ministry said Trillion Labs was also assessed as having strong technological self-reliance, including by demonstrating performance on major global leaderboards, but did not make the final selection.

Motif Technologies will receive support including 768 Nvidia B200 GPUs. It will also receive 1.75 billion won for building and processing individual datasets and 10 billion won for joint data purchasing and use. It will also be granted the title of a 'K-AI company'.

The second-stage evaluation will keep the existing framework of benchmark, expert and user assessments. A proposal is being reviewed to further subdivide a proprietary-technology assessment within the expert evaluation items.

Bae Kyung-hoon said major big tech companies such as OpenAI and Anthropic were not huge organisations from the start. He said the government would spare no policy support to keep the country's AI ecosystem alive.