Who ranks second in AI web coding, Anthropic or OpenAI?

A public leaderboard that tests AI models on building websites has just changed its order at the top, and a Polymarket contract is betting that the new ranking holds through September 30. The market asks which lab will be second at month-end, and it now prices OpenAI at 92% to finish in that spot.

The leaderboard comes from Arena, a site where visitors ask for a website and then vote on which of two anonymous AI models does a better job. Each lab is ranked by its strongest model, so the race for second place is a direct contest between Anthropic's best and OpenAI's best.

OpenAI's top model, GPT-6 Astra, led from September 5. Then Anthropic released Claude Opus 5.5 on September 22, and Arena added it to WebDev voting the same day. Anthropic had said Opus 5.5 performs at the level of its Claude Fable 5.1, which sits at 1,755 points. However, voters liked it much more. At the board's latest update, Opus 5.5 scored 1,818 to Astra's 1,792, putting Anthropic first and OpenAI second.

For OpenAI to lose second place, it would have to release a new model that beats 1,818 before the September 30 cutoff. Unfortunately, OpenAI's most recent releases, GPT-6 Sol and Luna, arrived on September 22 as cheaper GPT-6 variants, and Sol's best score of 1,686 sits well below both leaders.

The only remaining window is OpenAI's DevDay conference on September 29, the day before the market settles. Fortune reports that OpenAI may unveil GPT-6 Cyber there, a model built for cybersecurity work that only a small group of testers can use for now. Even if a surprise model appears, it would still need to reach the Arena board and post a score above 1,818 within a single day.

With OpenAI's latest models already scored and the third-place lab far behind, the path for OpenAI to move back into first is narrow. That is why the market's 92 cents on OpenAI finishing second makes sense, and we expect OpenAI to hold that spot on September 30.