Aya Expanse 32B is Cohere's multilingual research flagship — a 32B model trained specifically to serve the world's underrepresented languages with high-quality understanding and generation.
Aya Expanse 32B is an AI model released October 2024 by Cohere, a leading organization in large language model research. Cohere's multilingual specialist — state-of-the-art performance in 23 languages. Ideal for global content generation. With 32B and a 128K tokens context window, Aya Expanse 32B is designed to excel across a range of demanding tasks — from software engineering and mathematical problem-solving to multilingual content generation and long-document analysis.
Aya Expanse 32B uses a dense transformer architecture with 32B of parameters, all of which are active for every input token. Dense architectures provide predictable, consistent behavior across diverse task types — there are no "cold" experts to warm up, and quality is uniform across prompts. Cohere's training pipeline combined large-scale pretraining on Multilingual preference data across 23 languages from 35+ countries with supervised fine-tuning and alignment techniques (RLHF or DPO), producing Aya Expanse 32B's characteristic balance of capability, helpfulness, and safety.
As of 2026-10-16, Aya Expanse 32B by Cohere demonstrates measurable performance on standard AI evaluation benchmarks. On Multilingual MMLU — which tests world knowledge across 57 academic subjects from elementary to professional level — it achieves 77.4%. In mathematical reasoning, Aya Expanse 32B records 72.1% on MGSM. These scores place Aya Expanse 32B — at 32B of parameters — in a competitive position within its model class. Note that benchmark scores reflect controlled evaluation conditions; real-world performance varies by task, prompt design, and domain specificity.
The primary strengths of Aya Expanse 32B are: 23-language model — one of the most multilingual open-weight models; Purpose-built for cross-lingual understanding — not translated from English-only training; Apache 2.0-adjacent research license. These make it particularly well-suited for Global AI applications in 23+ languages, Multilingual customer service AI, NGOs serving diverse language communities. Users working on complex, multi-step problems benefit most from Aya Expanse 32B's ability to maintain coherence across long inputs and produce structured, verifiable outputs.
Like all current AI systems, Aya Expanse 32B has limitations: Research-only open access — commercial needs Cohere API; English performance below specialized English-only models. For production deployments, validating outputs against authoritative sources — especially for domain-specific or time-sensitive information — remains best practice. Cohere continues to improve Aya Expanse 32B through ongoing research.
Aya Expanse 32B is available on LITHOVEX AI with a free account. Unlike direct API access — which requires billing setup, API key management, and per-token charges — LITHOVEX AI provides a seamless browser interface with no setup required, making Aya Expanse 32B accessible to students, researchers, developers, and early-stage teams globally.
As of 2026-10-16, Aya Expanse 32B by Cohere demonstrates measurable performance on standard AI evaluation benchmarks. On Multilingual MMLU — which tests world knowledge across 57 academic subjects from elementary to professional level — it achieves 77.4%. In mathematical reasoning, Aya Expanse 32B records 72.1% on MGSM. These scores place Aya Expanse 32B — at 32B of parameters — in a competitive position within its model class. Note that benchmark scores reflect controlled evaluation conditions; real-world performance varies by task, prompt design, and domain specificity.
Dense transformer, 32B parameters, optimized for multilingual instruction following
The training pipeline for Aya Expanse 32B involves multiple stages: large-scale pretraining on Multilingual preference data across 23 languages from 35+ countries to develop broad world knowledge, followed by supervised fine-tuning on instruction-response pairs, and alignment techniques such as RLHF or DPO. Each stage refines the model's helpfulness, accuracy, and safety. Cohere's research team — based in Canada — has optimized Aya Expanse 32B for the specific capability profile described above.
Inference efficiency: Aya Expanse 32B at 32B is designed to deliver quality outputs at practical serving speeds. Its 128K tokens context window enables processing entire documents, multi-file codebases, and extended conversation histories — particularly valuable for contract review, technical documentation analysis, and codebase-level AI assistance.
Getting started with Aya Expanse 32B on LITHOVEX AI takes under two minutes. Follow these steps:
For best results with Aya Expanse 32B: provide specific, context-rich prompts; include examples when generating code; specify output format when structure matters; break complex tasks into focused sub-questions.
AI applications for African language communities
Southeast Asian multilingual customer service
South Asian regional language content creation
Arabic-language document analysis and generation
Research into underrepresented language AI capability
Cohere is a Toronto-based AI company founded in 2019 by researchers including Aidan Gomez — one of the co-authors of the original "Attention Is All You Need" transformer paper. The company focuses on enterprise-grade language models optimized for business: retrieval-augmented generation (RAG), tool use, and multilingual document processing. The Command model family supports up to 256K token context windows, 23+ languages, and reliable structured output — capabilities specifically tailored for financial services, healthcare, and legal enterprise customers. Cohere has raised over $440 million and counts Oracle, Salesforce, and NVIDIA among its investors.
Funding: $445M raised (NVIDIA, Oracle, Salesforce)
Founded: 2019, headquartered in Canada.
Aya Expanse 32B by Cohere competes in a field with several strong alternatives. Below is a detailed side-by-side comparison against commonly listed alternatives, using benchmark data and capability profiles in the catalog.
Both Aya Expanse 32B and the models above are available free on LITHOVEX AI — test them side by side on your real tasks before committing to a subscription with any provider.
Aya Expanse 32B is released under CC BY-NC 4.0, which means you can download and run it on your own hardware. Below are the quickest ways to get started.
LITHOVEX AI runs Aya Expanse 32B on cloud infrastructure, so you can use it instantly in your browser without downloading anything, configuring drivers, or managing VRAM. Create a free account and start in under 60 seconds.
Aya Expanse 32B is Cohere's most ambitious multilingual model, covering languages from Arabic and Swahili to Bengali and Tagalog that are poorly served by US-centric AI models. Critical for AI applications in Africa, South Asia, Southeast Asia, and the Middle East where linguistic diversity is the norm.
Aya Expanse 32B by Cohere has the following version history and notable updates:
For the complete changelog and latest release notes, visit the official Cohere model page.