**Fragmented token volumes across multiple high-performing models have left the #2 AI lab position on OpenRouter for the week of August 31–September 6 highly contested, with market-implied odds clustering tightly around 45–50% for leading contenders like Google and “Other.”** DeepSeek’s V4 Flash series leads overall volume through low-cost MoE architectures optimized for agentic and high-throughput workloads, but second place splits among Tencent’s Hy series, Z.ai’s GLM models, OpenAI’s GPT-5.6 variants, Xiaomi’s MiMo, MiniMax, and Qwen. Recent developer adoption favors aggressive Chinese pricing and open-weight releases for coding and tool-use tasks, while OpenAI discounts temporarily boosted share without displacing the broader shift toward cost-efficient alternatives. This dynamic underscores how platform usage now rewards inference economics and iteration speed over frontier capability alone, keeping resolution sensitive to the precise weekly aggregation.