Launch HN: Tokenless (YC S26) – Automatic Model Switching to Cut Costs

Tokenless is a drop-in replacement model router designed to cut costs on API calls.

Here's how it works:

• When a request comes in, it fans out the request to multiple models simultaneously (Fan-out).
• It monitors each model's reasoning process, and as soon as it determines that a particular model is close to the correct answer, it immediately selects that model and stops execution of the remaining models.
• This allows users to pay only for the tokens they actually need, maintaining quality without being forced to always use a frontier model.

Key features:

• Provides OpenAI- and Anthropic-compatible endpoints, minimizing changes to existing code.
• Rather than relying on simple marketing claims, it measures and provides solve rate and cost per task based on a public Agentic Benchmark as its performance metric.
• Users can simulate expected savings based on their monthly LLM spend.