How much funding has Motif raised?

Motif has raised Undisclosed in total funding. Their latest funding round was Undisclosed. Key investors include Government of Korea (National AI Foundation Model Project), Samil PwC (Strategic Partner).

Where is Motif headquartered?

Motif is headquartered in Seoul, South Korea. The company has a team of N/A people.

Motif operates in the AI Infrastructure sector. A South Korean AI startup developing high-performance open-source large language models (LLMs) and multimodal foundation models with proprietary Grouped Differential Attention (GDA) architecture for enterprise and personal AI applications. Delivers frontier-level AI performance at significantly lower cost through proprietary GDA architecture and full-stack optimization (infrastructure, model architecture, operations), enabling high-performance models to run under constrained compute budgets while matching or exceeding larger proprietary models.

When was Motif founded?

Motif was founded in 2025. It is based in Seoul, South Korea.

Who is the CEO of Motif?

Lim Jung-hwan (임정환) is the CEO of Motif. The company was founded in 2025.

Where is Motif located?

Motif is headquartered in Seoul, South Korea.

What funding stage is Motif at?

Motif is currently at the Commercial stage.

How many employees does Motif have?

Motif has approximately N/A employees.

Motif

Category: AI Infrastructure

A South Korean AI startup developing high-performance open-source large language models (LLMs) and multimodal foundation models with proprietary Grouped Differential Attention (GDA) architecture for enterprise and personal AI applications. Motif was founded in 2025. The company is led by Lim Jung-hwan (임정환). Based in Seoul, South Korea. Team size: N/A. Total funding raised: Undisclosed. Latest round: Undisclosed. Key investors include Government of Korea (National AI Foundation Model Project), Samil PwC (Strategic Partner).

AMW Analysis

Motif develops open-source large language models and multimodal foundation models using its proprietary Grouped Differential Attention architecture, targeting enterprise and personal AI applications. The South Korean startup, founded in 2025, has secured government backing through the National AI Foundation Model Project and a strategic partnership with Samil PwC. Its recent news flow shows rapid product iteration and growing investor confidence. In December 2025, a 12.7-billion-parameter model outperformed GPT-5.1 on reasoning benchmarks, emphasizing data alignment over scale. By February 2026, Motif-2-12.7B, built in seven weeks, surpassed larger models from Alibaba and Google on reasoning tasks. In May 2026, the company raised a $17 million Series B round from existing investors Nice Investment Partners and Nautilus Investment, joined by Dito Investment and Forest Ventures. Most recently, in July 2026, the 314-billion-parameter Motif-3 beta, developed under the government’s Independent AI Foundation Model project, tied DeepSeek V4 Pro with an AAII score of 44. The trajectory shows consistent performance gains across model sizes, with funding and government partnerships supporting continued development of efficient, high-performance foundation models.

AMW analysis, generated from 4 tracked news signals.

Founded: 2025
Headquarters: Seoul, South Korea
Team size: N/A
Total funding: Undisclosed

Value proposition

Delivers frontier-level AI performance at significantly lower cost through proprietary GDA architecture and full-stack optimization (infrastructure, model architecture, operations), enabling high-performance models to run under constrained compute budgets while matching or exceeding larger proprietary models.

Products and solutions

Motif-2-12.7B-Base, Motif-2-12.7B-Instruct, Motif-2-12.7B-Reasoning, Motif-2.6B Lightweight Model, Personal AI Platform (Custom Model Fine-Tuning), Enterprise AX Solutions, Multimodal AI Capabilities (Text-to-Video in Development)

Unique value

Only Korean startup with demonstrated experience developing both high-performance large language models (LLMs) and large multimodal models (LMMs) as foundation models. The Motif-2-12.7B-Reasoning model outperformed Alibaba's Qwen 2.5 (72B parameters) in mathematics, science, and logic tasks despite being significantly smaller (12.7B parameters). Open-sourced all models on Hugging Face platform, making advanced Korean-language AI capabilities publicly accessible.

Target customer

Enterprises seeking cost-efficient AI solutions, customer support teams building chatbots, marketing teams requiring branded content generation, AI development teams, startups, and individual creators

Industries served

Artificial Intelligence Infrastructure, Enterprise Software, Customer Support & Service, Marketing & Content Creation, AI Development & Research, Technology & Innovation

Technology advantage

Proprietary Grouped Differential Attention (GDA) architecture that disentangles signal and noise-control attention pathways, improving representational efficiency and enabling robust instruction generalization under constrained compute budgets. Full-stack optimization approach across GPU infrastructure, kernels, runtime, serving, and model architecture—similar to DeepSeek's methodology. Models trained from scratch on AMD Instinct™ MI250 GPUs, demonstrating hardware-agnostic efficiency. Fine-tuning capabilities with small datasets for enterprise customization without requiring large-scale internal data.

How they differentiate

Only Korean startup with demonstrated experience developing both high-performance large language models (LLMs) and large multimodal models (LMMs) as foundation models. Proprietary Grouped Differential Attention (GDA) architecture enables superior reasoning performance at significantly lower computational cost—Motif-2-12.7B-Reasoning outperformed Alibaba Qwen 2.5 (72B parameters) in mathematics, science, and logic tasks despite being 6x smaller. Open-source approach with all models publicly available on Hugging Face platform, combined with full-stack optimization across GPU infrastructure, kernels, runtime, and serving.

Main competitors

Trillion Labs, Upstage, Naver Cloud (HyperCLOVA)

Key partnerships

Samsung C-Lab (Global Partner for Enterprise AX Projects), Microsoft Azure (Cloud Infrastructure and Global Distribution), Samil PwC / Samil Accounting Corporation (Enterprise AI Transformation and Strategy), Korean Government - Independent AI Foundation Model Project (Elite Team Selected Feb 2026), AMD (GPU Infrastructure Partner for Model Training), Hugging Face (Open-Source Model Distribution Platform), Moreh (Parent Company - GPU Infrastructure and Distributed Inference), CrowdWorks (Data Construction for AI Models), Endartright (VLA Synthetic Data for Robotics), Seoul National University, KAIST, Hanyang University (Academic Research Partners)

Notable customers

Samil PwC (Enterprise AI Transformation Partnership)

Major milestones

Founded in February 2025 in Seoul, South Korea as a spin-off from Moreh, Developed Motif-2-12.7B-Reasoning model outperforming Alibaba Qwen 2.5 (72B) in math, science, and logic benchmarks (November 2025), Selected as one of 15 participants in Korea's National AI Foundation Model Project (2025), Open-sourced Motif-2-12.7B model family on Hugging Face platform (2025), Selected as elite team in Independent AI Foundation Model project, receiving 768 Nvidia B200 GPUs (February 2026), Featured in VentureBeat for enterprise LLM training innovations (December 2025), Partnership with Samil PwC for enterprise AI transformation initiatives, Developed proprietary Grouped Differential Attention (GDA) architecture, Models trained on AMD Instinct MI250 GPUs demonstrating hardware-agnostic optimization

Growth metrics

Selected as elite team (4th) in Korea's Independent AI Foundation Model Project (February 2026), receiving 768 Nvidia B200 GPUs and ~KRW 11.75 billion data budget. Competing alongside LG AI Research Institute, SK Telecom, and Upstage for final selection (2 of 4 teams) by end of 2026. Consortium of 17 organizations. Motif-2-12.7B ranked #1 Korean model and #11 globally on AAII benchmarks. Developed Motif-Image-6B and Motif-Video-1.9B multimodal models. 26 employees as of March 2026.

Market positioning

Agile startup challenger in Korea's sovereign AI initiative. Selected as the 4th elite team in the Ministry of Science and ICT's Independent AI Foundation Model Project (February 2026), joining LG AI Research Institute, SK Telecom, and Upstage. Competing for final selection (2 of 4 teams) by end of 2026. Consortium includes 17 organizations. Received 768 Nvidia B200 GPUs and ~KRW 11.75 billion in data budget. Positioned as a technically superior but resource-constrained alternative to conglomerate-backed teams.

Geographic focus

Primarily South Korea for sovereign AI initiatives and enterprise customers. Selected as elite team in government's Independent AI Foundation Model Project with consortium of 17 organizations. Strategic partnerships with Samsung C-Lab and Microsoft Azure enable potential global distribution. Hugging Face platform provides international accessibility to open-source models. Focused on Korean language and market needs as part of national AI sovereignty strategy, with roadmap expanding to 300B+ parameter models and physical AI (VLA).

Patents and IP

No registered patents publicly disclosed as of February 2026. Core innovations are open-sourced through technical reports and model releases on arXiv and Hugging Face.

About Lim Jung-hwan (임정환)

AI and machine learning expert; CEO leading Motif (also referred to as Motif Intelligence in Korean media) in developing large language models and AI infrastructure solutions. Pioneer in developing Korea's sovereign AI foundation models.