Cohere Labs

2026-09-16

MODELS · 1 SOURCE · Cohere

Cohere's Tiny Aya expedition yields multilingual community projects and COLM 2026 paper

Cohere Labs' open-weight multilingual model Tiny Aya was distributed through 'Expedition Tiny Aya', producing community research and prototypes across education, multilingual safety, on-device deployment, and language understanding. Projects include a Math Edition, an offline multilingual Kids Companion, and safety analyses of code-mixed prompts; related research from the initiative was accepted to COLM 2026.

7.0

2026-09-16

RESEARCH · 1 SOURCE · Cohere

Cohere Labs preprint finds a ‘culture funnel’ in LLM pipelines and publishes CultureMarkers dataset

Cohere Labs analyzed over 5.6 million training samples across pretraining, SFT, alignment and reasoning datasets and reports a consistent pattern — a ‘culture funnel’ where cultural diversity narrows as data moves into post-training stages. The team used Cohere’s Command A model to tag cultural signals, argues that multilingual coverage alone doesn’t ensure cultural representation, and published a preprint on arXiv plus the CultureMarkers dataset on Hugging Face to support further study.

6.0

2026-09-16

RESEARCH · 1 SOURCE · Cohere

Cohere Labs publishes Agentic Task Ecosystem (ATE)—~696K MCP tools aggregated into a dataset

Cohere Labs aggregated seven public AI-tool directories to create the Agentic Task Ecosystem (ATE), a corpus of roughly 696,000 published tools across 123,000 MCP servers. Under a strict test of whether a tool can actually carry out an occupational task, only 2.6% of tools qualify; 419 of 923 U.S. occupations show no agentic-tool activity, and patterns show tools mainly (1) represent existing work at finer grain, (2) provide agent infrastructure, or (3) create a small amount of new work (mostly agent management), with expert judgments of technical feasibility predicting which occupations receive tools.

8.0

2026-09-16

RESEARCH · 1 SOURCE · Cohere

Cohere Labs: Limitations of 2023 'GPTs are GPTs' exposure scores are shaping policy

Cohere Labs argues that the widely cited 2023 paper 'GPTs are GPTs'—which estimated how many U.S. jobs have tasks exposed to LLMs—was a bounded technical exercise tied to a 2023 GPT-4-era model and a U.S. occupational taxonomy, yet its exposure scores are now being used by the IMF, OECD, U.S. Senate and others to inform 2026 policy decisions. The brief warns these limitations compound when scores travel across models, countries, and non-itemizable work, and calls for more dynamic, representative measurement tools and worker-centered research to better inform policymakers.

7.0