Cohere details Expedition Tiny Aya results

// 1h agoNEWS

Cohere details Expedition Tiny Aya results

Cohere For AI showcased the results of Expedition Tiny Aya, a global research initiative using the 3.35B-parameter multilingual Tiny Aya model family for edge devices. Highlighted projects include child-safe offline voice assistants, Android tool-calling systems, and multilingual math benchmarks.

// ANALYSIS

While tech giants race to build larger frontier models, Cohere For AI's Expedition Tiny Aya proves that small, highly optimized models are the true path to democratizing AI globally, especially for low-resource languages and on-device use.

–Compact Efficiency: 3.35B parameter models enable real-world deployment on edge and mobile devices, drastically lowering barriers to entry.
–Multilingual First: Rather than treated as an afterthought, low-resource languages are first-class citizens, filling a massive gap left by English-centric models.
–Community Collaboration: Shifting research from isolated labs to global open-source sprints accelerates practical applications like educational assistants and local tool-calling.

// TAGS

cohereexpedition-tiny-ayatiny-ayamultilingual-aiopen-sourceartificial-intelligenceedge-computing

DISCOVERED

1h ago

2026-07-15

PUBLISHED

2h ago

2026-07-15

RELEVANCE

7/ 10

AUTHOR

cohere

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

UPDATE37m ago

Vercel CLI v55.0.0 accelerates project linking

Vercel CLI v55.0.0 revamps the project linking flow by resolving team scopes first, restricting searches to the resolved team instead of sweeping all associated accounts. In non-interactive CI and agent environments, the CLI now fails early with a structured JSON error if a scope is missing and multiple teams exist.

RESEARCH1h ago

RoboTTT scales robot memory to 8K

Stanford and NVIDIA have introduced RoboTTT, a test-time training framework that allows robots to dynamically adapt and learn during deployment rather than relying on frozen pre-trained weights. By reframing inference as a continuous self-supervised learning problem, the model scales context length to 8,000 timesteps with constant latency.

OPEN SOURCE1h ago

xAI open-sources Grok Build coding agent

xAI engineer skcd42 announced the open-sourcing of the complete Grok Build application code. Developers can now inspect the inner workings of the terminal-based AI coding agent, including its terminal rendering engine, agent loop execution, context compaction algorithms, goal tracking, and subagent execution systems.