The model race shifts to agent systems

Polymarket’s AI leaderboard betting shows the monthly best-model race has become volatile and inconclusive. In October 2026, Google led Anthropic 71-26 for the month, while the year-end contract showed Google 42%, Anthropic 42%, and OpenAI 9%, suggesting traders see model leadership as temporary rather than defensible.

Meta’s Muse Wonder launch reinforced that pressure, arriving at $4 per million input tokens and $20 per million output tokens, described as 60% cheaper than Fable 5.1 and GPT-6 Astra. Industry commentary argues that technical moats are shrinking, with OpenRouter usage data showing 84% of tokens going to non-state-of-the-art models and token usage up roughly 50% since July while prices fell 41%.

Durable advantage is framed around agent systems, proprietary interaction data, and distribution. The agent market has Anthropic at 93%, OpenAI at 4%, and Google at 3%, pointing to the value of orchestration layers, workflow integration, memory, tool routing, recovery paths, and accumulated user context rather than the underlying model alone.