Welcome Fireworks.ai on the Hub 🎆

Source: HuggingFace Date: 2025-02-14 URL: https://huggingface.co/blog/fireworks-ai

Summary

Integration announcement: Fireworks.ai added as a HF Inference Provider, enabling serverless inference for DeepSeek-R1/V3, Llama 3.2 90B Vision, Qwen2.5-Coder, and others directly from HF model pages. No HF markup on top of Fireworks pricing. HF PRO subscribers get $2/month in credits. Access via InferenceClient(provider="fireworks-ai") through HF’s router endpoint.

Implications

Thread: HF as open-source ML hub. Another inference provider added to the HF router expands the competitive options for users who want HF as the unified access layer. Fireworks.ai is known for high-throughput optimized inference, which matters for DeepSeek-R1 and similar large models that need efficient batching. The no-markup pricing model through HF’s router is notable — HF is subsidizing inference access with PRO credits to keep users in the HF ecosystem rather than going directly to providers. This is a retention play as much as a capability expansion.