dots.llm1
142B-total, 14B-active MoE from RedNote's hi lab, trained without synthetic data and said to match Qwen2.5-72B; intermediate checkpoints also released.
A data-pipeline-focused release, with pretraining on high-quality non-synthetic corpora and open checkpoints across the full training run for research on learning dynamics. MIT.
- Date
- Friday, 6 June 2025
- Lab
- Xiaohongshu (RedNote)
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Total / active parameters | 142B / 14B | company |
Release date 2025.06.06 from the model card news line. Comparison to Qwen2.5-72B is the authors' claim.
Sources
This record was checked against its sources on 6 October 2026. How we check