The closely matched odds around 20-28% for leaders like Google and Z.ai reflect intense competition among AI labs to climb Text Arena math leaderboards by late November, driven by rapid iteration on reasoning models. Google benefits from Gemini's established math benchmarks and scale, while Z.ai and peers like DeepSeek or Moonshot gain from specialized training on math datasets and efficient inference advances. Key differentiators include access to high-quality synthetic data, integration of chain-of-thought techniques, and hardware optimizations that boost performance on GSM8K-style tasks without proportional compute increases. Upcoming model updates and benchmark releases could shift the field quickly given the narrow gaps.