Inference token spend has been shifting away from closed models since mid-July, and the convergence runs both directions.

SiliconData figures via the Financial Times show enterprise spend moving toward open models over the past month, though absolute proprietary spend stays far higher. The wrinkle: effective prices for proprietary models fell sharply as OpenAI cut rates, while effective prices for open models actually rose, as stronger Chinese near-frontier models like GLM-5.2 and Kimi K3 got served at premium rates. TechRepublic That undercuts the simple "open is the cheap option" framing.