As of November 17, 2026, Anthropic holds the top spot on the Artificial Analysis Intelligence Index. Its Fable 5.2 model scores 64.6, ahead of OpenAI’s GPT-6.2 Astra at 62.1. That gap has persisted through seven consecutive months of Anthropic leading, its longest unbroken run since the two companies started trading the top position.
The Benchmark
The Artificial Analysis Intelligence Index currently combines nine tests covering knowledge work, coding, science, and general knowledge. Artificial Analysis updated its methodology on October 27, 2026, adding two tests, upgrading three, and removing three. Scores before and after that date are not directly comparable.
How the Lead Changed Hands
Between September 2024 and April 2026, leadership changed hands eight times. Some margins were thin and short-lived. Anthropic’s Opus 4.5 held the lead for just 17 days in late 2025 before OpenAI’s GPT-5.2 overtook it in December.
The streak Anthropic is on now began in April 2026, when Opus 4.7 scored 40.69 and pushed ahead of GPT-5.4’s 38.98. Since then, neither a methodology change nor a rapid OpenAI response has closed the gap. GPT-6 Astra arrived two days after Fable 5.1 and scored 0.7 points lower. GPT-6.2 Astra followed Fable 5.2 by eight days and finished 2.4 points behind.
Since March 2024, Anthropic has spent 529 days at the top of the index versus 459 for OpenAI.
Release Pace Is Accelerating
In 2025, the two companies released a new frontier model roughly every 52 days. So far in 2026 through mid-November, that interval has compressed to about 25 days, with 13 total releases. Anthropic accounts for eight of those 13.
Benchmark Leadership Is Not the Full Picture
On Epoch AI’s Capabilities Index, OpenAI’s GPT-6.2 Astra leads with a score of 174.2, narrowly ahead of Fable 5.2 at 173.5. The difference falls within both models’ margins of error.
On market share, ChatGPT held 88% of global AI chatbot website traffic at the end of 2024. By September 2026 that share had fallen to 56% as Claude and Google Gemini gained ground. Winning benchmarks and winning users are two different scoreboards.
