Same Model, 10x the Price: Why Stripe Bought the Routing Layer

Stripe has agreed to acquire OpenRouter, a platform that routes AI model calls across more than 400 models from over 80 providers based on cost, speed, and task complexity. The deal bundles two decisions agents have always made separately into a single infrastructure layer: which model to use and how to pay for it.

The pricing spread is what makes routing matter. OpenRouter data from June showed Llama 3.3 70B input priced at $0.10 per million tokens through DeepInfra and $1.04 through Together AI, a 10x difference for the same underlying model. Output pricing ranged from $0.32 to $1.04 per million tokens across providers. That gap is not a rounding error. For an agent running thousands of calls per day, routing intelligently across providers is a cost strategy, not a convenience feature.

OpenRouter handles two routing decisions: which model handles a request, and which provider endpoint serves it. The platform evaluates task complexity, price, speed, and reliability in real time. It also provides failover when a provider hits rate limits, context-length errors, or goes down.

This is already live infrastructure for a significant share of the market. F5's State of Application Strategy report, based on responses from more than 1,100 IT decision-makers, found that 52% of organisations were already chaining or orchestrating multiple AI models, with respondents using an average of seven models each.

Stripe brings the payment layer. The combination means the routing layer, the part of your stack that decides which model-provider pair handles each call, is now the same entity that bills for it. That creates a clear set of new products: per-task pricing tied to model selection, spend optimization across providers, consolidated billing for multi-model workflows.

Snowflake announced its own dynamic model routing for Cortex AI Gateway on August 18. The routing layer is no longer a niche developer tool. It is becoming the payment layer for agent infrastructure.

For builders running multi-model workflows today, three configuration decisions matter before the Stripe acquisition closes and pricing structures shift. First, set task-complexity thresholds to gate expensive model usage: route cheap tasks to cheaper endpoints and keep frontier models for inference that actually requires them. Second, set per-call price caps so a provider surge does not blow your budget during high-volume runs. Third, if any of your prompts touch sensitive data, flag those paths for Zero Data Retention endpoints only. OpenRouter supports all three today.

The agent economy is not just about what models can do. It is about what they cost and who controls the routing. Both of those just moved to the same company.