Back to Gap Map Model components / Training & synthetic datasets

Dolma 3

open / Overall score: 4.0(strong)

Ai2's pretraining data release behind OLMo 3: a pool of about 9 trillion tokens drawn from Common Crawl, olmOCR-processed scientific PDFs, Stack-Edu code and FineMath, alongside the quality-upsampled mixes used in training - Dolma 3 Mix, Dolmino mid-training and Longmino long-context. The pool, every mix and the curation tooling are released together.

A family record covering the pool and the training mixes; successor to the map's Dolma 1.x entry.

Openness

5 high confidence

5.0

license: ODC-BY
gated: false
dataset_card: present

ODC-BY, ungated, fully documented pool and training mixes.

https://huggingface.co/api/datasets/allenai/dolma3_pool recorded 2026-09-29
Hub API JSON for allenai/dolma3_pool has the gated field set to false (not gated) and tags "license:odc-by", "size_categories:10B<n<100B" and "arxiv:2512.13961".
https://huggingface.co/datasets/allenai/dolma3_pool recorded 2026-09-29
HF dataset page for allenai/dolma3_pool renders the dataset card, stating "Dolma 3 is licensed under the Open Data Commons Attribution License v1.0 (ODC-By)" and "The Dolma 3 pool is a dataset of over 9 trillion tokens".
https://huggingface.co/datasets/allenai/dolma3_pool/raw/main/README.md recorded 2026-09-29
Raw dataset card front matter lists "license: odc-by" and the body states "Dolma 3 is licensed under the Open Data Commons Attribution License v1.0 (ODC-By)" with a documentation overview of the pool, its sources and loading instructions.

Adoption

3 high confidence

3.0

Adoption is measured by Hugging Face downloads of the dataset over the trailing month, the only usage channel on record for Dolma 3.

https://huggingface.co/api/datasets/allenai/dolma3_pool recorded 2026-08-13
36,467 downloads in the trailing 30 days for allenai/dolma3_pool

Capability

5 high confidence

5.0

The current (2025) corpus behind OLMo 3, the strongest fully-open base and thinking models; uses documented quality-aware upsampling and data-mix ablations.

https://allenai.org/blog/olmo3 recorded 2026-09-29
Ai2 Olmo 3 blog states "Together, Dolma 3 and Dolci give Olmo 3 a fully open data curriculum from first token to final post-trained checkpoint" and "Olmo 3-Base 32B outperforms other fully open base models".
https://arxiv.org/abs/2512.13961 recorded 2026-09-29
arXiv abstract for the Olmo 3 paper says "We introduce Olmo 3, a family of state-of-the-art, fully-open language models at the 7B and 32B parameter scales" and "Our flagship model, Olmo 3 Think 32B, is the strongest fully-open thinking model released to-date".

Verified 2026-09-27