VIDRAFT Open Discovery Challenge 2026: Benchmarking LLMs
South Korean AI firm VIDRAFT launches a Hugging Face challenge for LLMs to design malaria and TB drugs, drawing 2,000+ submissions in three days.
Stock photo for illustration only, not from the actual event
- VIDRAFT launched the Open Discovery Challenge on Hugging Face for AI drug discovery.
- Focuses on malaria and tuberculosis to address market failures in low-income populations.
- Racked up over 2,000 candidate submissions within just three days of launch.
- Supports diverse AI models including OpenAI, Claude, Gemini, DeepSeek, and Qwen.
The intersection of artificial intelligence and pharmaceutical research has reached a new milestone as South Korean deep-tech AI and science research company VIDRAFT introduces the Open Discovery Challenge. Hosted publicly on Hugging Face, the competition benchmarks large language models on their ability to propose novel molecular structures specifically targeted at treating malaria and tuberculosis, evaluating candidates against real pharmacological criteria rather than static multiple-choice questions.
Both diseases were chosen deliberately due to their massive public health burdens in low-income regions, where pharmaceutical market incentives are typically weak. To bridge this gap, VIDRAFT opened the competition to developers, researchers, and general enthusiasts alike, allowing participants to leverage any available AI systems—ranging from commercial powerhouses like OpenAI, Claude, Gemini, and DeepSeek to proprietary or open-weight models like Qwen and Kimi.
Submissions are funneled through a comprehensive multi-factor scoring pipeline that measures efficacy potential, toxicity, target binding, ADME characteristics, and preclinical or clinical simulations. Early data collected within the first 72 hours of launch already highlights measurable performance differences across various major LLM families when tackling authentic scientific workflows.
Stock photo for illustration only, not from the actual event
From a structural standpoint, this challenge reframes drug discovery as an evaluation of complete "science agent" stacks rather than raw language models alone. Success relies heavily on prompt engineering, tool integration, and research strategy, allowing non-specialists to meaningfully contribute by following the dedicated AI drug development guide provided within the challenge space.
Operating as a living dataset, the challenge is expected to evolve into a continuously updated benchmark for general AI scientific reasoning. Interested researchers can access the competition space directly through the official VIDRAFT organization page on Hugging Face for the latest guidelines and participation instructions.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment