Recent agentic workloads and rapid adoption of low-cost open-weight large language models from Chinese labs such as DeepSeek V4.1 Flash and GLM 5.3 Flash continue to drive OpenRouter token volumes, with open-weight models handling roughly 72% of traffic. Recent weekly totals have stabilized near 165–184T amid new model releases, high cache-hit rates on agent prompts, and sustained developer routing through the platform. These factors create tight clustering around the 175–205T range, as volume leaders remain distinct from higher-priced closed models like Claude Opus 5.5 that dominate revenue. Any acceleration from fresh releases or shifts in agent usage patterns could tilt outcomes within the narrow probability bands.