Which company owns the best AI model on September 30?

Google released Gemini 3.7 Flash on August 13, three weeks after its last model, and it went straight to sixth place on the public rankings. The five models above it are all made by Anthropic. Polymarket's best AI model market predicts a 94% probability that this will still be true on September 30, which means betting on Anthropic provides a 6.38% return in 30 days.

The market settles on one web page at noon on the last day of the month. It is arena.ai, where a person types a question, gets two anonymous answers, picks the better one, and only then finds out which model wrote which. Nearly 8 million of those votes have been turned into a score.

The version this market reads has style control switched off. So, a long, heavily formatted answer is not penalized for being long and heavily formatted. On that setting, Anthropic's Claude Opus 5 Max sits first on 1504.7 points, and the next four places are Anthropic too. The best model anybody else owns is 14.4 points behind.

Anthropic has held all top 5 positions for 5 months now. What changed each time was simply which new Anthropic model was better than the last.

So can anyone ship something in September that closes 14.4 points in 30 days? Quite unlikely. No lab has said it will. Google is training Gemini 4, and Sundar Pichai has pointed to the end of the year. xAI's Grok 5 has missed three announced windows. OpenAI's newest flagship, GPT-5.6, arrived in June and ranks 36th.

Sure, new models do arrive without warning, tested anonymously under codenames before launch, and there are unnamed entries on the site right now belonging to xAI, Google, Alibaba, and Meta. But none belong to a lab that has been within 14 points of first place this year.