Featherless.ai

Featherless.ai provides serverless AI inference, letting developers run open models without managing the underlying infrastructure. As of August 2026, its catalog offers 40,000+ models through one API key, supporting uses such as coding, reasoning, multilingual applications and creative writing. Individuals can use an interactive chat subscription, developers can build applications with usage-based inference, and businesses can choose dedicated GPUs with engineering support. The company also offers a managed runtime for open-source agents, starting with OpenClaw.

The company’s co-founders are CEO Eugene Cheah, CTO Harrison Vanderbyl and Wesley George. Originally operating as Recursal.AI, it adopted the Featherless.AI identity in 2025. Its research includes RADLADS, a method for converting pretrained transformers to alternative attention architectures without training from scratch. The process aligns hidden states, distills model outputs and fine-tunes for long contexts; Featherless used it to train the attention-free QRWKV-72B model with eight GPUs. This research complements its work on inference, model and workflow optimization.

In April 2026, the company reported more than 10,000 customers and monthly revenue above $250,000 but below $500,000—an annualized revenue run rate above $3 million. It announced a $20 million Series A that month, co-led by AMD Ventures and Airbus Ventures, to expand infrastructure, model access and enterprise deployments.