SYNTH
PleIAsLarge synthetic dataset from PleIAs, released with the AI Alliance, designed for training small reasoning models end to end. It holds tens of millions of samples amplifying content from tens of thousands of Wikipedia articles, Wikibooks and internal documents, across exercise types including arithmetic, creative writing, RAG and information extraction. Samples carry reasoning traces and seed attribution, and roughly a fifth covers non-English European languages.
Openness
5 high confidence- license
- CC-BY-4.0(permissive)
- access
- ungated
- dataset_card
- present
Permissively licensed, ungated dataset with a documented card.
- https://huggingface.co/api/datasets/PleIAs/SYNTH recorded 2026-09-29
Hub API record for the dataset lists the license tag as "license:cc-by-4.0" and the card title "SYNTH - generalist open data and environment"; its gated field is false, so downloads need no access request.
- https://huggingface.co/datasets/PleIAs/SYNTH recorded 2026-09-29
Dataset page renders the card, including "SYNTH is the first open generalist synthetic dataset" and the license "cc-by-4.0".
- https://huggingface.co/datasets/PleIAs/SYNTH/raw/main/README.md recorded 2026-09-29
Raw card front matter sets "license: cc-by-4.0" and the card describes itself as "is the first open generalist synthetic dataset for training small reasoning model end-to-end".
Adoption
3 medium confidenceAdoption is measured by Hugging Face downloads of the dataset over the trailing month, the only usage channel on record for SYNTH. The count sits near a boundary, so an ordinary month of variation could move the reading either way.
- https://huggingface.co/api/datasets/PleIAs/SYNTH recorded 2026-08-14
downloads field = 10,111 in the trailing 30 days; 273 likes; gated false
Capability
4 medium confidencePleIAs trained Baguettotron (321M parameters) and Monad (56M) end to end on SYNTH and reports Baguettotron as best in class among models under 350M. The evidence comes from a single lab, with no independent evaluation.
- https://huggingface.co/blog/Pclanglais/synth-data-frontier recorded 2026-09-29
PleIAs post states "Baguettotron is currently the best in class across major industry benchmarks (MMLU, gsm8k, HotPotQA)" and that "We trained two reasoners on SYNTH".
- https://huggingface.co/datasets/PleIAs/SYNTH/raw/main/README.md recorded 2026-09-29
Card claims "SYNTH includes 79,648,272 individual text samples" and "achieving current best results for size range on MMLU and other standard evaluation metrics".
Verified 2026-09-27