# Hugging Face Inference Is Now Powered by OpenRouter

## Executive summary

Hugging Face has partnered with OpenRouter to route all inference requests through OpenRouter, providing users with access to various model providers directly from Hugging Face model pages. This integration allows users to view detailed provider metrics (such as price per million tokens and tokens per second) and manage usage, billing, and specific provider access within the Hugging Face ecosystem.

## Key takeaways

- Unified Inference Routing: Hugging Face requests are now routed via OpenRouter to ensure the 'best possible inference' is available, regardless of the underlying provider.
- Advanced Model Metrics: The platform displays detailed metrics for selected models, including price per million tokens (input/output), tokens per second, time to first token, and context window.
- Coding Agent Integration: Models can be utilized directly within various coding agents, including Pi coding agent, Open Code, Claude Code, and Codex.
- Provider Control and Filtering: Users can manage inference providers in settings, allowing them to filter out or disable specific providers (e.g., Novita, Cerebras, Fal) from automatic routing.

## Technical details

- Model and Provider Specifics: The DeepSeek-V4.1-Flash-1017 checkpoint is cited as a current example of an open-weight model available through the system.
- Billing and Credits: Hugging Face Pro accounts include $2 monthly credits, which are stated to be sufficient for approximately 150 million tokens of DeepSeek-V4.1-Flash, assuming prompt caching.
- Pricing Transparency: The displayed prices are guaranteed to have no markup and match the prices announced by OpenRouter.

## Practical implications

- Engineers can streamline LLM usage by accessing multiple providers through a single interface (Hugging Face).
- Cost management is enhanced through the ability to monitor usage analytics and set spending limits for enterprise accounts.
- The integration supports advanced development workflows by enabling direct API calls in Python and JavaScript, and integration with various coding agents.

## Topics

LLM Inference, API Integration, Resource Management, OpenAI Alternatives, AI Tooling, Hugging Face, OpenRouter, DeepSeek-V4.1-Flash-1017

Source: https://www.youtube.com/watch?v=TmZ4RCXYsTQ
