hf-inference: hf CLI extension for Hugging Face Inference Providers

Run inference on thousands of models through Hugging Face Inference Providers directly from the command line.

Installation

curl -LsSf https://hf.co/cli/install.sh | bash
hf extensions install hf-inference

Usage

Run inference

hf inference run "What is the capital of France?" --model Qwen/Qwen3.5-35B-A3B
hf inference run "Explain quicksort" --model moonshotai/Kimi-K2.5 --stream
hf inference run "Translate to French: hello world" --model Qwen/Qwen3.5-35B-A3B --provider cheapest

Pick a specific provider or routing policy with --provider:

hf inference run "Hello" --model moonshotai/Kimi-K2.5 --provider novita
hf inference run "Hello" --model Qwen/Qwen3.5-35B-A3B --provider cheapest
hf inference run "Hello" --model Qwen/Qwen3.5-35B-A3B --provider fastest

Pipe input via stdin:

cat article.txt | hf inference run --model Qwen/Qwen3.5-35B-A3B --system-prompt "Summarize this"

List available models

hf inference list
hf inference list --provider novita
hf inference list --search qwen -n 5
hf inference list --format json
hf inference list -q                       # model IDs only

Show provider details for a model

hf inference info moonshotai/Kimi-K2.5

PROVIDER      STATUS  CONTEXT  INPUT $/M  OUTPUT $/M  TOOLS  STRUCTURED
------------  ------  -------  ---------  ----------  -----  ----------
fireworks-ai  live    262144                          yes    no        
novita        live    262144   0.6        3           yes    no        
together      live    262144   0.5        2.8         yes    no   

Environment Variables

Variable	Purpose
HF_TOKEN	Hugging Face API token (also reads from ~/.cache/huggingface/token)