VIDRAFT, a Korean AI startup, has released Darwin-180B-RSI, a 180-billion-parameter mixture-of-experts model. This model is built upon Qwen3.8-Flash-Next and incorporates selective model merging with a recursive self-improvement technique. VIDRAFT claims Darwin-180B-RSI has achieved top rankings on five Hugging Face leaderboards across various reasoning benchmarks, though these results are self-reported and lack independent validation. AI
IMPACT This model's claimed performance on reasoning benchmarks could influence future development in MoE architectures and self-improvement techniques.
RANK_REASON The cluster describes a new model release with technical details and benchmark claims, but lacks direct announcement from a frontier lab. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
