Launch HN: Tokenless (YC S26) – Automatic model switching to save money
usetokenless.com
July 29, 2026
2 min read
🔥🔥🔥🔥🔥
46/100
Summary
Tokenless is a router that reduces inference costs by half by directing API calls to a group of models and selecting the most promising one. It offers an OpenAI and Anthropic compatible endpoint for easy integration.
Key Takeaways
- Tokenless is a router that reduces inference costs by up to 50% by optimizing API calls to the most suitable models.
- The service provides a compatible endpoint for OpenAI and Anthropic models, allowing easy integration for users.
- Tokenless achieves an average cost per task of $0.57, significantly lower than other leading models.
- The company was founded by AI researchers from Google DeepMind, Princeton, and UC Berkeley and is backed by Y Combinator.
Community Sentiment
Positives
- The multi-model progress monitoring idea is clever — querying in parallel and deciding mid-turn feels like a game changer for dynamic routing.
- Having a composable architecture for confidence prediction and routing makes it easier to adapt to new models without retraining from scratch.
- The approach of cutting off poorly performing model responses to save costs is a smart move that could lead to significant savings in token usage.
Concerns
- There's skepticism about the real intelligence behind model selection — if it's just a classifier, the system might be slow and expensive.
- Concerns arise about the economic viability of switching models, especially when most agentic work involves long tool-call chains that benefit from a hot cache.
- The need to retrain routing rules with every new model release raises questions about the long-term sustainability of this approach.
