Cohere on Hugging Face Inference Providers 🔥

Source: HuggingFace Date: 2025-04-16 URL: https://huggingface.co/blog/inference-providers-cohere

Summary

Integration announcement: Cohere becomes the first model creator to serve its own models as an official HF Inference Provider. Nine models available including Command A (111B, 256k context), Aya Expanse (32B, 23 languages), Aya Vision (32B multimodal), and Command R7B. Features: RAG with citations, tool use, multilingual (23+ languages), long context. Access via HF UI, huggingface_hub SDK, JS client, or OpenAI-compatible endpoint. Standard Cohere pricing; no HF markup on routed requests.

Implications

HF as open-source ML hub. Cohere serving through HF as the first model creator (as distinct from compute provider) in the Inference Providers program is a notable category expansion. Previously, providers like Together AI and Groq ran open-weights models on their own hardware; Cohere bringing its proprietary models through HF’s routing layer positions HF as an inference marketplace that spans both open-weights and commercial proprietary models.

Open-weights ecosystem health. Aya Expanse and Aya Vision being available via HF Inference Providers gives the multilingual open community access to strong non-English models through the standard SDK interface. Aya’s 23-language coverage is particularly relevant for teams needing multilingual instruction following that English-dominant open-weights models don’t reliably provide.