Skip to main content
Rase AI Intelligence Newsletter

Models & assistants Model or AI service

Aya Expanse brings multilingual text models to 23 languages

Cohere released Aya Expanse in 8B and 32B sizes for multilingual writing, translation and research. The downloadable weights use non-commercial terms. [1][3]

Model access
8B and 32B weights are downloadable. The API catalog lists 32B as live and 8B as retired. [2][3]
License
CC BY-NC with Cohere Labs’ acceptable-use policy for the documented weights. [3]

What the model does

Cohere’s research team introduced Aya Expanse on October 24, 2024. It released two instruction-tuned checkpoints and published evaluations across a set of 23 languages. [1][3]

Why it matters

The models make it possible to compare results in multiple languages with the same checkpoint. That is useful when a project combines language tasks that are usually tested separately. [1][3]

Who can use it

Researchers can download the two sizes to compare multilingual tasks. Cohere continues to list the 32B model in its API catalog, while the hosted 8B model retired. [2][3]

Key dates

  1. Aya Expanse weights introduced [1]
  2. Hosted 8B API model retires [2]

Ways to use it

For multilingual document work, evaluate each target language against representative inputs. Compare translation or summary accuracy on regional wording alongside the provider’s general benchmarks. [1][3]

Model limitations

The weights use Attribution-NonCommercial terms and Cohere Labs’ acceptable-use policy. The provider’s language list describes its evaluated coverage, and output quality can vary by language and task. [2][3]

Sources and verification

Catalog evidence checked How we verify records

  1. Primary source Cohere’s Aya Expanse launch announcement
  2. Primary source Cohere API model catalog, including Aya Expanse availability
  3. Primary source CohereLabs/aya-expanse-32b
  4. Primary source Hugging Face registry metadata for CohereLabs/aya-expanse-32b

Related profiles

Models & assistants

GPT-6.2 Astra and Sol

OpenAI

Astra targets demanding projects. Sol targets everyday tasks. Both support text and image input in the API.

High confidenceEvidence checked

Models & assistants

MAI Code 2 and Code 2 Flash

Microsoft AI

Code 2 targets complex software work. Flash targets frequent edits, explanations and routine fixes.

High confidenceEvidence checked

Models & assistants

MAI Realtime 1

Microsoft AI

Users can try a MAI Labs demo. Developers can register for a limited Foundry preview.

High confidenceEvidence checked