Google and Anthropic currently lead Text Arena Math rankings with top models scoring 1530 and 1522 respectively, reflecting strong performance on competition-level benchmarks like AIME and FrontierMath. Trader sentiment remains closely balanced because both labs continue rapid iteration on reasoning architectures, while OpenAI’s September 2026 claims of solving over 100 open problems and the Navier-Stokes Millennium Prize have not yet translated into arena dominance. Moonshot, Z.ai, DeepSeek, and Meta trail but post competitive scores and lower costs, creating uncertainty about whether specialized math training or multi-agent systems will pull ahead by year-end. Upcoming model releases and benchmark updates before December represent the main swing factors.