RWKV

RWKV is the open-source model project around which Recursor originally built its product in Featherless AI: When Your Weekend Experiment Makes More Than Your Startup. Eugene Chia describes it as a next-generation foundation model direction with the potential for much cheaper inference.

The episode’s main RWKV lesson is not a technical evaluation of the architecture. It is a pivot lesson: a model can fit a mission of cheaper and broader AI access, while the market still pulls harder toward already demanded model families such as Llama, [[MistralAI|Mistral]], Qwen, and DeepSeek.

Connections

• Recursor - company/product direction built around RWKV before the pivot.
• Featherless AI - later platform that moved from own-model focus toward many open models.
• Open Source AI Models - broader model ecosystem RWKV belongs to in the source.
• AI Inference Cost Structure - cost motive behind the original RWKV bet.
• Customer Pull - reason the team shifted toward mainstream open models.