Darwin-180B-RSI-R3: Top Open-Weight Model on SuperCLUE November 2026, Outscoring Gemini and DeepSeek in Math & Science
Darwin-180B-RSI-R3: Top Open-Weight Model on SuperCLUE November 2026, Outscoring Gemini and DeepSeek in Math & Science TL;DR: VIDRAFT's open-weight model Darwin-180B-RSI-R3 scored 73.68 in math and science on the November 2026 SuperCLUE benchmark — #1 among all open-weight models and #2 overall out of 19 models. Built on top of Qwen3.8-Flash-Next and fine-tuned via a Recursive Self-Improvement…
VIDRAFT has unveiled its open-weight model, Darwin-180B-RSI-R3, which has claimed the top spot on the November 2026 SuperCLUE benchmark for math and science. The model outperformed both Google's Gemini-4-Argon and DeepSeek-V4.1-Flash, scoring 73.68 in math and science, ranking #1 among open-weight models and #2 overall out of 19 models.
Built on Qwen3.8-Flash-Next and fine-tuned using VIDRAFT's proprietary Recursive Self-Improvement (RSI) training method, Darwin-180B-RSI-R3 is available for public use on Hugging Face and has a quantized variant for consumer hardware. The model's innovative RSI approach involves iterative self-refinement, generating solutions, verifying them, and fine-tuning the model based on correct reasoning traces.
While the overall SuperCLUE score is 9th among 19 models, Darwin-180B-RSI-R3's strength lies in its exceptional mathematical reasoning abilities.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
