OpenRouter’s Alex Atallah and AMP’s Anjney Midha discuss the company’s evolution from open-weight models to a multi-model routing layer used by over 10 million developers

OpenRouter’s origin story centers on a bet made when most observers expected only one or two frontier model labs to emerge. The company’s early focus on open-weight models like Llama, Alpaca, and Mistral positioned it as a neutral layer for developers seeking to route between multiple providers rather than rely on a single vendor.

The company’s evolution into a routing layer was not initially obvious to investors, who at times dismissed OpenRouter as “just a wrapper” or “just a marketplace.” Despite this skepticism, OpenRouter became a critical piece of infrastructure for more than 10 million developers, enabling them to switch between models without rewriting integrations. This shift reflected a broader industry move away from closed, single-model ecosystems toward competitive, multi-model environments.

OpenRouter’s platform now processes over 10 trillion tokens per day, a scale that reflects its role as a real-time map of AI usage. Its leaderboards and rankings have become a live snapshot of how developers and users are engaging with different models, capturing shifts in preference and capability that are otherwise difficult to observe. The company’s focus on distribution—helping model labs overcome the challenge of getting their checkpoints into developers’ hands—distinguished it from competitors that prioritized training or fine-tuning.

The podcast highlights OpenRouter’s early experiments with model fusion, including a first iteration that was later deleted and reintroduced years later. These efforts were part of a broader exploration of how to combine models dynamically, though the company ultimately chose to focus narrowly on routing rather than expanding into adjacent areas like fine-tuning or memory. This strategic focus became one of its biggest advantages as the inference marketplace matured.

Stripe’s acquisition of OpenRouter is framed as a strategic fit, driven in part by shared concerns about token fraud in the emerging AI economy. As token flows become more valuable and autonomous agents proliferate, the risk of fraud shifts from human actors to automated systems. OpenRouter’s infrastructure, which sits at the intersection of model routing and token economics, positions it to address these challenges alongside Stripe’s existing fraud prevention capabilities.