Motif 3 Beta: Korea's 314B MoE model challenges DeepSeek and MiniMax
A 13B active parameter MoE model from a Korean startup goes head-to-head with giants.
Motif Technologies has released Motif 3 Beta, a sparse Mixture-of-Experts (MoE) language model with 314 billion total parameters and only 13 billion active per token. This architecture allows the model to maintain a large knowledge base while keeping inference compute costs low, similar to approaches used by DeepSeek V4 Pro (1.6T total, 49B active) and MiniMax-M3 (428B total, 23B active). The model is available on Hugging Face under the Motif-Technologies organization, making it accessible for researchers and developers to test.
Motif 3 Beta is not just a technical release—it is a strategic move in South Korea's national AI Foundation Model Project, which invests ₩530 billion ($0.36B) through 2027. After an initial evaluation round in December 2025, Naver Cloud and NC AI were dropped, and Motif Technologies was selected as an additional funded company in February 2026. The next evaluation in August 2026 will determine which of the remaining five companies—Upstage, SKT, LG AI Research, Motif, and the original five minus the dropped ones—continue to receive government support. This project underscores South Korea's ambition to build sovereign AI capability, and Motif 3 Beta is a key entry in that race.
- Motif 3 Beta uses a sparse MoE architecture: 314B total parameters, only 13B active per token, enabling efficient inference.
- Competes against DeepSeek V4 Pro (1.6T/49B active) and MiniMax-M3 (428B/23B active), positioning as a mid-range active parameter model.
- Motif Technologies joined South Korea's AI Foundation Model Project with ₩530B ($0.36B) investment after Naver Cloud and NC AI were eliminated in the first evaluation round.
Why It Matters
A 13B active MoE model from a Korean startup signals growing global competition in efficient large-scale AI.