The Korea Economic Daily Reports Darwin First Among Open-Source Models in SuperCLUE Math and Science Reasoning
aillmbenchmarknews

TL;DR
On November 16, 2026, The Korea Economic Daily (Hankyung), Korea's leading business daily, reported that VIDRAFT's large language model Darwin-180B-RSI ranked first among open-source (open-weight) models in the mathematical and scientific reasoning track of SuperCLUE's November 2026 evaluation. Including closed models, Darwin placed second, directly behind OpenAI's GPT-6.1-Sol.

What The Korea Economic Daily reported
| Point | What the article says |
|---|---|
| Result | First among open-source models and second overall in math and science reasoning |
| How SuperCLUE grades | The evaluator grades each model itself every month on private questions. Developers do not submit scores |
| Size | A 180 billion parameter model scoring more than 10 points above a 2.8 trillion parameter model |
| Inference speed | 18,057 tokens per second on server GPUs, and 2.6 times the concurrent throughput of a commercial 4-bit model |
| Korean NPUs | 417 tokens per second on an NPU, and VKORE, the technology for training and running large models on Korean NPUs |
| On-device | POCKET, the light edition that runs on a laptop and keeps data inside the device |
| Downloads | About 5 million cumulative downloads on Hugging Face and more than 1.2 million a month |
| ModelScope | The only Korean AI organization registered as an officially verified organization |
| Hugging Face records | Number one in 12 official benchmark leaderboards, GPQA 94.44, full marks on AIME 2026 and HMMT |
| Method | Synthesis of the strengths of parent models, then model-level recursive self-improvement on self-generated solutions with automatic answer checking |
| Safety | ViKernel, VIDRAFT's safety technology that judges and blocks risky model actions in advance, used in a national security foundation model project |
The article closes with CEO Kim Minsik's comment that Korea can lead the efficiency race, where a smaller model delivers more performance.
Why this coverage matters
The Korea Economic Daily is the business newspaper that Korean executives and investors read. Its coverage follows Seoul Shinmun on November 13 and Chosun Biz, Economist Korea, Digital Daily and AI Times on November 6. The article places the SuperCLUE result next to the business picture: inference speed, Korean semiconductors, on-device deployment and download numbers.
The numbers
Mathematical and scientific reasoning, SuperCLUE November 2026:
| Rank | Model | Developer | Open weights | Score |
|---|---|---|---|---|
| 1 | GPT-6.1-Sol | OpenAI | No | 76.32 |
| 2 | Darwin-180B-RSI-R3 | VIDRAFT | Yes | 73.68 |
| 3 | Gemini-4-Argon | No | 71.05 | |
| 3 | DeepSeek-V4.1-Flash | DeepSeek | Yes | 71.05 |
| 5 | Grok-4.7 | xAI | No | 68.42 |
Links
- The Korea Economic Daily article (Korean): https://www.hankyung.com/article/202611168298O
- SuperCLUE board: https://www.superclueai.com/homepage
- Model: https://huggingface.co/FINAL-Bench/Darwin-180B-RSI-R3
- Laptop edition: https://huggingface.co/FINAL-Bench/POCKET-Darwin-180B-GGUF
- Hugging Face official benchmark leaderboards map: https://huggingface.co/spaces/quantid/huggingface-official-benchmark-leaderboards
- VIDRAFT press list: https://vidraft.net