VIDRAFT's Darwin-180B-RSI model family has achieved the top position in 9 official Hugging Face benchmarks, surpassing all other participating organizations. This open-source model utilizes a recursive self-improvement technique, where it solves verifiable problems, retains only its correct solutions, and retrains itself without human-generated answers. Notably, it excels in practical tasks such as structured output generation (IFStruct) and data extraction from PDFs (ExtractBench), demonstrating its real-world applicability. AI
IMPACT Sets a new standard for open-source model performance on structured data tasks and benchmarks.
RANK_REASON The cluster details a new open-source model release and its performance on benchmarks, including a novel training technique.
- Darwin-180B-RSI
- ExtractBench
- GPQA: A Graduate-Level Google-Proof Q&A Benchmark
- Hugging Face
- IFStruct
- Liquid Ai
- MMLU-Pro
- Moonshot AI
- Qwen3.8-Flash-Next
- VIDRAFT
- Zhipu AI
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →