Which company has the best AI model end of September?

The board has Anthropic near 98.7%, and the frontier data mostly backs it. Anthropic has held the top of the leaderboard since February, and Claude Opus 5.5 shipped September 22, leading several coding harnesses the last week of the month, with Fable 5.1 anchoring knowledge-work benchmarks. The catch is how close OpenAI's GPT-6 Astra runs, tying Fable 5.1 on one re-based intelligence index and winning a couple of the sub-benchmarks. Whoever writes the resolution criteria decides this, since 'best' depends on which index and which snapshot date get used. A late Google or OpenAI release would test the price, but the calendar is nearly out.