10x Tool Calls in 12 Months: How to Drive Agent Success with High-Quality Inference with OpenRouter's Co-Founder and COO

Today's best frontier model might not be next month's best, and the leaderboard reshuffles fast enough that any single-provider bet can go stale before your annual contract is up. The labs know this. Their sales motions are built to lock in large commits early, securing your traffic before you've had time to benchmark the next release. OpenRouter processes 2T tokens daily across 300+ models and 60+ providers. That gives us a unique window into how inference is bought and consumed at scale, and how model and provider optionality unlocks your ability to optimize for intelligence, cost, performance, and fit-to-task across your AI workflows. In this session, Chris Clark will share what that dataset reveals: a framework any AI leader can apply to buy inference that maximizes AI capability. You'll learn why inference is uniquely hard to buy, how commits with providers, clouds, and labs actually work, what to realistically expect in terms of discounts, and how to preserve optionality without leaving money on the table. You'll see how to evaluate models across intelligence, cost, performance, and fit-to-task rather than just price per token, and where open-weight models are production-ready today. And you'll walk away understanding which architecture decisions impact your inference quality and cost, things like caching and model and provider routing.