AI Potluck
Model components / Training & synthetic datasets

Aya Collection

Cohere

Multilingual instruction-tuning dataset spanning more than 114 languages, built from templated and translated subsets for tasks including classification, summarization and translation. It holds hundreds of millions of examples across 64 subsets, and was created by Cohere Labs as part of the Aya open-science initiative.

Verified 2026-08-13 via the Aya Collection dataset card on Hugging Face.

Openness

5 high confidence
5.0
license
apache-2.0
gated
false
card
present

Ungated, Apache-2.0 licensed, with a complete dataset card.

Adoption

3 high confidence
3.0

11,190 downloads in the trailing 30 days for CohereLabs/aya_collection, which bands at 10K-100K, level 3 on the dataset adoption scale.

Capability

4 high confidence
4.0

The instruction collection behind Aya 101 (2024), Aya 23 and Aya Expanse, which beat Gemma 2, Qwen 2.5 and Llama 3.1 in their size class: the Aya Expanse paper reports win rates up to 76.6%, reached by the 32B against Mixtral 8x22B on m-ArenaHard, and the accompanying blog post reports the 8B range at 60.4% to 70.6% against Gemma 2 9B, Llama 3.1 8B and Ministral 8B. Either comparison set is open models this collection fed, so the band holds on both readings. The weakest link is the Aya Expanse lineage itself: no primary source names the Aya Collection as its training data, and the attribution rests on the secondary write-up cited here.

  • https://arxiv.org/abs/2412.04261 recorded 2026-08-14

    Aya Expanse paper abstract - "Aya Expanse 8B and 32B outperform leading open-weight models in their respective parameter classes, including Gemma 2, Qwen 2.5, and Llama 3.1, achieving up to a 76.6% win-rate" on Arena-Hard-Auto translated into 23 languages, and "Aya Expanse 32B outperforms Llama 3.1 70B ... achieving a 54.0% win-rate"

  • https://huggingface.co/blog/aya-expanse recorded 2026-07-04

    Reports the Aya Expanse 8B only, with "win rates ranging from 60.4% to 70.6%" against Gemma 2 9B, Llama 3.1 8B and Ministral 8B

  • https://ritvik19.medium.com/papers-explained-151-aya-23-d01605c3ee80 recorded 2026-07-04

    Aya Collection (513M examples, 114+ langs) feeds Aya 101 and Aya 23/Expanse instruction data

Verified 2026-08-13