Anthropic holds overwhelming market-implied odds on LiveBench Mathematics at end of September due to its Claude Opus and Fable series models posting top scores in the benchmark’s math category, including standout results near 97% on recent evaluations of numerical reasoning and problem-solving tasks. These outcomes reflect consistent advantages in handling complex, contamination-resistant questions that test genuine mathematical capability over pattern matching. Traders appear to view Anthropic’s latest releases and iterative improvements as establishing a durable edge in this specific domain relative to OpenAI, Google, and other labs. A late-month model update from a competitor or a benchmark refresh introducing harder problems could still narrow the gap before resolution.