上海创智学院与新序智能开源预训练RSI模型801-30B
Start here
2026年11月4日，上海创智学院与新序智能联合发布首份公开的预训练级递归自我改进（RSI）技术报告，并开源了801-30B基座模型及轻量SFT数据。该技术方案将不可逆的预训练大循环拆解为调试、潜力和经验量化三个可闭合的小循环，利用多智能体在超参数、代码与数据管线间排查异常。评测表现方面，801-30B基座模型在13项基础能力评测中平均得分73.13，并在轻量微调后展现出较强的代码与智能体评测表现。
AI-generated from reports · Updated 19 hours ago
Report timeline
Follow the reports to see different sides of the story.
- AITNT · AI头条SelectedShanghai Chuangzhi Institute and Xinxiu Intelligence Release Pre-trained RSI Model 801-30B and Technical Report
Shanghai Chuangzhi Institute and Xinxiu Intelligence jointly released the first public technical report on pre-training-level Recursive Self-Improvement (RSI), open-sourcing the 801-30B base model and lightweight SFT data. The framework decomposes the irreversible pre-training macro-loop into three closeable micro-loops: debugging, potential, and empirical quantification, leveraging multi-agent systems to detect anomalies across hyperparameters, code, and data pipelines. Across 13 foundational capability benchmarks, the 801-30B base model averaged a score of 73.13 and demonstrated strong coding and agent benchmark performance following lightweight fine-tuning.
Story heat trend
Current heat 6·Comparable peak 10（Nov 4 16:00）·Comparable change, last 24 hours–
Trends compare only the same parties observed continuously, so the scope can be narrower than current heat counts. Move the pointer or tap the chart to see hourly heat; use the left and right arrow keys on a keyboard.