Spokes.wiki Search Graph Growth About

llm-providers-wiki

Organization updated Mon Jun 15 2026 00:00:00 GMT+0000 (Coordinated Universal Time)

Cohere

Canadian AI lab (Toronto, founded 2019) built for enterprises, not consumers. No public chatbot product. Its position in the provider landscape is sovereign AI: private cloud, on-premise, and regional deployments (EU, APAC, UK) for regulated industries that can’t route data through shared cloud infrastructure.

Background

Founded by Aidan Gomez (co-author of the original Transformer paper), Ivan Zhang, and Nick Frosst. Cohere’s early bet was “same architecture, different deployment model” — enterprise API and on-premise contracts rather than a consumer product competing with ChatGPT.

Model families

Command series — instruction-following and RAG-optimized: Command R and Command R+ (retrieval-tuned, long context); Command A (later generation).

North family — newer generation; includes North Mini Code (cohere-north-mini-code), Cohere’s first agentic coding model — and, notably, Apache-2.0 open-weight (30B MoE / 3B active, on Hugging Face). So the North family is where Cohere stepped onto the open-weight axis, not just a developer-facing extension of the closed enterprise base.

Embed and Rerank — retrieval infrastructure: text embeddings and search reranking for enterprise RAG stacks. Available via amazon-bedrock alongside Cohere’s own API.

Position in the market

Cohere fills the compliance/sovereignty slot the spoke’s map otherwise lacked: regulated industries (finance, healthcare, government) that need data residency guarantees — the kind of requirement that disqualifies shared-cloud APIs from the other labs. The distinctive axis is the deployment model (private cloud / on-premise / regional), not consumer frontier reach (openai, anthropic).

But sovereignty and open weights are not opposite axes — cohere-north-mini-code shows Cohere pursuing sovereignty through Apache-2.0 open weights you can run on your own H100, which puts it alongside deepseek, qwen, and llama on the open-weight axis rather than across from it. North Mini Code also opens a developer-coding front (vs. GitHub Copilot and code-tuned open models), so Cohere now runs two plays at once: sovereign enterprise deployment and open-weight developer tools.

llm-provider · open-weight-models · amazon-bedrock · cohere-north-mini-code