On February 17, 2026, Cohere Labs, the research arm of enterprise AI company Cohere, released Tiny Aya — a family of open-weight multilingual models that covers more than 70 languages in 3.35 billion parameters, small enough to run on a laptop. TechCrunch covered the launch, which landed on the sidelines of the India AI Impact Summit in New Delhi.
Why it matters: multilingual capability has concentrated in frontier models sitting behind APIs, and quality degrades fast once you leave high-resource languages. Tiny Aya pushes the other way — open weights, deliberate emphasis on low-resource languages, and local deployment as the design goal, at a size consumer hardware can actually run. For teams building outside the English-first internet, that combination has not existed at this price point: free.
70+ Languages in 3.35B Parameters
The base model is 3.35 billion parameters, with BF16 weights totaling roughly 6.7 GB. Language coverage spans European, Indic, East and Southeast Asian, African, and Middle Eastern families — 67 languages tagged on the Hugging Face model card, with deliberate emphasis on low-resource ones: Zulu, Wolof, Igbo, Lao, Javanese, and Burmese all make the list. That emphasis is the point. High-resource languages are already served by every frontier API; the languages on this list are the ones where API models quietly fall back to translation pipelines and second-rate generation quality.
Architecturally, the model mixes sliding-window attention (a 4096 window with RoPE position encoding) with one global-attention layer, at 8K input and output context. That hybrid is the standard compromise when a small parameter budget has to serve long multilingual documents: sliding windows keep compute predictable, while the single global layer preserves enough long-range signal for cross-sentence coherence in languages the model sees less of during training.
TinyAya-Global: The Instruction-Tuned Variant
TechCrunch singled out TinyAya-Global, a variant fine-tuned to follow user instructions for applications that need broad language support. Per the model card, it is an alignment of tiny-aya-base through supervised fine-tuning plus preference training, positioned as the family’s best balance across languages and regions.
It Really Fits on a Laptop: Deployment
The model card states the design goal plainly: local deployment under realistic compute constraints. The practical paths run through transformers, vLLM, SGLang, and Docker Model Runner, and the community has already published eight quantizations. If you would rather not self-host, Cohere’s hosted inference serves roughly 133 tokens per second — useful for prototyping before you commit to running it yourself. Set against frontier models that weigh hundreds of gigabytes, this size turns “offline multilingual assistant” from a demo into a shippable product.
The deployment story matters more than the raw parameter count. A 6.7 GB BF16 model fits on ordinary laptops and small cloud instances; with the published quantizations it goes lower still. That is the difference between a model you evaluate and a model you actually ship to users in markets where bandwidth and privacy expectations both argue for on-device inference.
License and Capability Limits
The license is CC-BY-NC-4.0 (non-commercial) plus Cohere’s acceptable use policy; the repository is gated behind a contact form, and commercial use means talking to Cohere sales. The capability limits are stated honestly: the card lists a single eval, GPQA Diamond at 28.28% — far below frontier reasoning models — and openly notes weaker chain-of-thought and math performance (MGSM-class tasks) plus the risk of factual errors, especially in low-resource languages. The positioning is clear — a workhorse for generation, translation, and cross-lingual summarization, not a reasoning model. If your product needs multi-step planning, keep a frontier model in the loop and use Tiny Aya where the job is language coverage at zero marginal cost.
The Aya Lineage and What Builders Should Take
Tiny Aya extends Cohere’s Aya program, an open-science multilingual initiative whose earlier research model covered 101 languages. Debuting at the India AI Impact Summit was a deliberate stage: the markets that need cheap, multilingual, local AI most are exactly the Global South users the summit gathered.
Two takeaways for product teams. First, unit economics in non-English markets change: a 3.35B local multilingual base means zero per-token bills and data that never leaves the device — cost and privacy improve at the same time. Second, the non-commercial license is the startup red line: prototype freely, but negotiate before you ship — the most common last mile for open-weights products.
Sources
- Cohere launches a family of open multilingual models — TechCrunch
- Cohere Labs Launches Tiny Aya for Multilingual AI — Cohere
- Tiny Aya Global — Hugging Face
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.