Tokenless Launches API Gateway for Dynamic LLM Routing, Halving Agent Costs

Tokenless, an API gateway developed by Rohit, Andrew, and Kev, dynamically routes LLM agent traffic between different models to reduce operational costs. The service targets developers building AI agents who face increasing token expenses from frontier models.

The core proposition of Tokenless is to deliver high-performance agent interactions at a lower price point. It achieves this by directing complex requests to more capable, often more expensive, frontier models only when necessary. Simpler turns are handled by more economical open-source alternatives.

The founders note that companies are exceeding their allocated AI budgets, highlighting a critical need for cost efficiency in LLM deployment. Tokenless addresses this by automating the model selection process that developers currently manage manually.

According to the team, their routing technique involves querying multiple models simultaneously and using their progress to inform routing decisions. This method is claimed to be novel and contributes to its efficiency.

"matches the performance of Claude Fable 5 at half the cost."

The company states it has developed a router capable of matching the performance of models like Claude Fable 5, but at half the cost. This suggests a significant efficiency gain for agent developers.

Tokenless offers $20 in free credit for new users to test the service. Future plans include integrating additional models such as Kimi K3 and other GPT variants, expanding its routing capabilities. The service appears to be US-based, given the founder's academic background.

For a Tokyo-based developer building AI agents, Tokenless could offer a direct path to more cost-effective operations. It streamlines the complex decision-making around model choice, allowing for more aggressive deployment of agentic workflows without prohibitive expense. The efficiency gains could make advanced AI agents more viable for smaller teams or projects with tighter budgets.