Arena becomes model routing infrastructure

This move turns Arena from a media like ranking surface into a toll collector on live inference spend. An evaluation product gets purchased around launches and bakeoffs. A routing API sits in the request path every time an app calls a model, decides whether to send traffic to the best, fastest, or cheapest option, and can keep earning as usage grows inside production software.

• The product shift is from advice to execution. Instead of showing a developer that model A beats model B, Arena can receive the API call itself through an OpenAI compatible endpoint, apply routing rules around quality, cost, and latency, then return the answer from the chosen model.
  1 sacra 2 sacra
• The closest comparable is OpenRouter. Its router chooses across many providers, tracks latency and throughput over rolling windows, supports failover, and monetizes by sitting between apps and model vendors. That is the infrastructure business Arena is moving toward, with recurring volume instead of episodic study fees.
  3 sacra 4 openrouter 5 openrouter
• This also creates a data flywheel. Once Arena routes real traffic, it sees which model actually works for coding, agents, and other production tasks under real cost and speed constraints. That is more valuable than static leaderboard votes because it links model choice to live application behavior.
  1 sacra 6 openrouter

If Arena keeps pushing traffic through Max and the Arena API, its advantage becomes operational judgment, not just benchmark credibility. The winning position is to become the default control layer between applications and a fragmented model market, where every new model release creates more routing complexity and more reasons to hand model selection to infrastructure.