Tokenless, a new API gateway from YC S26 founders Rohit, Andrew, and Kev, offers dynamic, turn-by-turn routing between AI models to significantly reduce costs.

Companies are struggling with escalating AI expenses, often exceeding budgets due to the high cost of frontier models. While open-source models are improving and cheaper, they don't yet match the capabilities of top-tier models. Tokenless bridges this gap by intelligently routing complex tasks to advanced models only when necessary, thereby lowering overall AI spend.

Developed by AI researchers, including former Princeton PhD student Rohit, Tokenless employs a novel technique that queries multiple models simultaneously and uses their progress to make routing decisions. This approach, detailed in their blog, allows them to match the performance of models like Claude Fable 5 at half the cost. They offer an OpenAI and Anthropic-compatible endpoint and provide $20 in free credit for users to test the service. Tokenless aims to provide frontier-level intelligence at a reduced price point.