Motif 3 API early access is open!Try it free

Models

Motif 3

NEW

A model built for agentic work

Reasoning, code, and tool use for multi-step, long-context tasks.

  • AAII score 47 (v4.1.1, Aug 2026)
  • Global rank 3
  • MIT license
Hugging FaceTech Report
Total parameters
314B
Active parameters per token
13.2B
Maximum context length
256Ktokens
License
MIT

  • Carry work across multiple steps

    Select tools, use their results, and move to the next step. Build AI agents that carry out complex tasks.

  • From understanding code to solving problems

    Support development tasks that require code changes and testing. Evaluated on solving issues in real repositories.

  • Connect long documents in one context

    Supports up to 256K tokens of context. Connect and synthesize information distributed across long documents.

Motif 2.6B

A compact model for reasoning and coding.

A 2.6B language model supporting Korean and English. Its compact size supports mathematical and logical problem-solving and code generation.

  • Compact model
  • Reduced hallucinations
  • Mathematical and logical reasoning
Hugging FaceTech Report
Parameters
2.6B
Training data
2.4Ttokens
Coding · HumanEval
68.3
Math · GSM8K
80.21

Motif Image 6B

From ideas in words to finely detailed images.

Turns scenes and moods described in text into images. Describe what you want in Korean and create images for your content.

  • Clarity powered by proprietary technology
  • Optimized for Korean prompts
  • Up to 2K resolution
Hugging Face

Generated images

Explore images generated with Motif Image 6B.

  • Welsh corgi on mint
    Welsh corgi on mint
  • Pastel townhouses
    Pastel townhouses
  • Portrait on lilac
    Portrait on lilac
  • Sunset over coastal cliffs
    Sunset over coastal cliffs

Motif Video 2B

A compact model with competitive video generation.

Designed for performance beyond model size. Separate structures handle prompt alignment, motion consistency, and detail to improve video generation. Supports both text and image inputs.

  • Text- and image-to-video
  • VBench score 83.76
  • Up to 720p resolution
Hugging FaceTech Report
Parameters
2B
VBench total
83.76
Maximum resolution
720p
Maximum frames
121frames

Generated videos

Watch the video and explore the prompt behind it.

View full showcase

Prompt: A sports car carving a sunset mountain corner

A sleek, midnight blue sports car with gleaming chrome accents approaches a sharp corner on a winding mountain road, the sun setting in the background casting a golden hue over the scene. The car's headlights pierce through the twilight, illuminating the path ahead. As it begins to turn, the tires grip the asphalt with precision, the vehicle's body leaning gracefully into the curve. The surrounding landscape blurs slightly, emphasizing the car's speed and agility. Dust kicks up from the road, creating a dramatic effect as the car completes the turn, the engine's roar echoing through the serene mountain pass.

Full-stack AI, built in house from infrastructure to models

  1. 01

    GPU INFRASTRUCTURE

    GPU infrastructure

    A vendor-neutral stack using both AMD and NVIDIA. We have built and operated large-scale AI infrastructure on AMD MI250.

  2. 02

    AI INFRA OPTIMIZATION

    Training and inference optimization

    We build training and inference middleware and operate the full stack.

  3. 03

    FOUNDATION MODEL

    Foundation models

    We develop our own LLMs and multimodal models from scratch.

  4. 04

    AI APPLICATION

    Services and agents

    We build and operate agent platforms and AI services tailored to businesses.

Bring AI into your work

We review the right model and deployment approach with you, based on your goals, data, and operating environment.

Talk to us