# VIDRAFT (비드래프트)

> Korea's independent Pre-AGI model developer. A small, elite team that takes on giant capital with just 24 GPUs. Home of the Darwin model family, the AETHER architecture, QuantumOS quantum cryptanalysis, and the FINAL-Bench metacognition benchmark. Website: https://vidraft.net

## About
VIDRAFT (Korean: 비드래프트, 非draft — "the undrafted") is a Korean AI company (part of the GINIGEN AI ecosystem) developing frontier, Pre-AGI–level models. The philosophy: intelligence is not built in one giant step — it is *grown* across generations, the way nature evolves. Founding principle: **do not exaggerate; state only what has been verified** — real trust comes from honestly naming what has NOT been done.

## Darwin — Model Family
Darwin is VIDRAFT's flagship LLM family, grown by crossbreeding the strengths of different models into a stronger next generation.
- **Catalog**: 227 models tracked (official VIDraft + FINAL-Bench releases plus community derivatives) with 1,000,000+ cumulative downloads on Hugging Face.
- **GPQA Diamond (global reasoning)**: multiple Darwin models rank among the world's top — e.g. Darwin-28B-Opus 88.9, Darwin-36B-Opus 88.4, Darwin-27B-Opus 86.9, Darwin-31B-Opus 85.9 — a rare feat for a single non-Chinese company.
- **Sizes**: from on-device 2B–4B up to 398B (JGOS), spanning dense and Mixture-of-Experts designs.
- Catalog & live dashboard: https://vidraft.net (see "모델" / Models page)
- Hugging Face org: https://huggingface.co/VIDraft

## K-AI Leaderboard (Korean-language reasoning)
On Korea's national K-AI Leaderboard (operated by NIA / Ministry of Science and ICT, 2026-04-28), **3 of the Top 5 models were Darwin derivatives** — Rogue-27B-KR (#2), AWAXIS-Think-28B (#4), and TenOS-Ko-28B-v3 (#5). Evaluated across 6 Korean dimensions (KMMLU-Pro, CLIcK, HLE-Ko, MuSR-Ko, Com2-main).

## AETHER — Pre-AGI Architecture
A next-generation architecture that looks beyond the Transformer: instead of a single attention mechanism, multiple kinds of attention (5·7·11) coexist and harmonize, with an explicit metacognition layer for self-doubt and self-correction. Positioned at the "Pre" tier on the FINAL-Bench AGI scale (Partial < Proto < **Pre** < Pass < Post).

## FINAL-Bench — Metacognition Benchmark
The world's first metacognitive benchmark for AI. Scale: 100 frontier tasks × 15 domains × metacognition types × difficulty levels. Key finding: declarative knowledge ≠ procedural self-correction (models that "know" often fail to catch their own errors).
- Global rank: Hugging Face Dataset #5, "Space of the Week".
- Dataset: https://huggingface.co/datasets/FINAL-Bench/FINAL-Bench
- Leaderboard: https://huggingface.co/spaces/FINAL-Bench/Leaderboard

## QuantumOS — Quantum Cryptanalysis (results only)
Genuine quantum-algorithm attacks on classical cipher structures, validated on **real IBM quantum hardware** (not simulation). Publicly reported milestone: exponential-structure recovery at N=10 on real hardware — beyond the known public record (N=4, to our knowledge).
Honest scope (stated plainly): this is **not** quantum supremacy, **not** a break of real AES/RSA or full 16-round DES, and world-first status is not yet peer-reviewed. Methodology is intentionally not disclosed; results are.
- Blog: https://huggingface.co/blog/FINAL-Bench/quantum
- Live demo: https://vidraft-quantumos.hf.space/crypto

## Drug & Materials Discovery
- **Polaris leaderboards**: 14 first-place finishes across ADMET, potency, and kinase-selectivity tasks (drug discovery).
- ADMET / toxicity models (PXR, Tox21) and materials-generation pipelines with real-hardware quantum cross-validation.

## Achievements & Recognition
- Selected for Korean government advanced-GPU support ("Blackwell GPU") program, 2026-04.
- Invited to Sapporo, Japan under the STEP (Sapporo Transnational Expansion Partnership) program, 2026-06 (Sapporo City Hall & Hokkaido Prefectural Government).
- Seoul AI Hub resident company; HuggingFace STAR AI Top-12 (only Korean company selected).
- Infrastructure: 24 GPUs (B200×16 + H200×8).

## Products & Demos
- Homepage / integrated dashboard: https://vidraft.net
- Real-time talking AI avatar: https://avatar.vidraft.net
- QuantumOS demo: https://vidraft-quantumos.hf.space

## Links
- Website: https://vidraft.net
- Hugging Face (VIDraft): https://huggingface.co/VIDraft
- Hugging Face (FINAL-Bench): https://huggingface.co/FINAL-Bench
- Quantum blog: https://huggingface.co/blog/FINAL-Bench/quantum
- Metacognition blog: https://huggingface.co/blog/FINAL-Bench/metacognitive

## Contact
VIDRAFT · GINIGEN AI · Seoul, Korea · https://vidraft.net