Featherless AI

Featherless is a serverless AI model provider that offers subscribers access to a wide range of Hugging Face models, with over 450 models available and new ones added weekly. It offers two pricing plans: Feather Basic, which allows access to models up to 15B in size for $10 per month, and Feather Premium, which provides access to models up to 72B in size for $25 per month.

Key features include:

• Unlimited personal use: Users can access an ever-growing list of models as long as they are subscribed.
• OpenAI compatible API: Users can access models via a familiar OpenAI API.
• Privacy-focused: Featherless does not store any user chat logs, prompts, or completions.
• FP8 quantization: Featherless uses FP8 quantization to maintain output quality while improving inference speeds.
• Dynamic model swapping: Featherless uses a custom inference stack that can swap models in less than a second, allowing for rapid reconfiguration and autoscaling.

Featherless offers advantages over using Hugging Face or other providers directly, such as lower cost, faster inference speeds, and a wider selection of models. The service also includes a referral program where users can get discounts for referring new subscribers.