Anthropic's Claude lineup, including recent Opus and Fable variants, has posted the strongest verified results on LiveBench's coding category through mid-October, driving the 87.5% implied probability. These models lead or place near the top in coding-specific evaluations that emphasize contamination-resistant tasks such as code generation, debugging, and multi-step agentic workflows, outpacing OpenAI's GPT-5 series and others in the latest public snapshots. Traders appear to view Anthropic's consistent edge in this benchmark subcategory—bolstered by iterative high-effort releases and strong third-party coding metrics—as durable through month-end. In contrast, OpenAI, Google, and Meta sit at low single-digit odds, reflecting comparatively weaker recent coding-specific gains on the live leaderboard despite competitive overall scores. No major new model drops or benchmark reversals have shifted sentiment in the past week.