~3.4B (Qwen2-family decoder-only Transformer)
Apache-2.0
Open-weight 3B base LLM release positioned as a transparency-first pretraining project. Official model-card claims emphasize full reproducibility artifacts beyond final weights, including training trajectories, intermediate checkpoints, data-processing decisions, and 200+ ablation studies across data quality, mixture design, training dynamics, and evaluation validity. Release materials describe a two-stage ~8T-token curriculum (6T broad web-scale pretraining + 2T structured QA/reasoning data) and claim an overall 51.72 average across 19 benchmarks, reported as approaching or matching some 7B-class baselines such as OLMo-3 7B. The card also frames the model as suitable for research on pretraining science and as a general-purpose base for language, reasoning, and code tasks, while explicitly noting it is a base model that is not instruction- or safety-aligned.