State of AI: 2026 AI Report — Research

stateof.ai [1] has produced the 2026 annual AI report [2]. The report is lead-authored by Nathan Benaich, a partner at Air Street Capital, and has been published for nine consecutive years since 2018, with peer review from leading AI labs, startups, policy circles, and academia. The report reviews AI progress over the past 12 months in Research, Industry, Politics, Safety, and other areas, and continues to publicly test the previous year's predictions. This article is translated and compiled from the report, with brief analysis added; reading the original report is strongly recommended.

The full table of contents follows. Owing to length constraints, this report will be published in three parts: the first covers Research, the second Industry, and the third combines Politics, Safety, and Predictions.

12 months pass, and the frontier fight is now a three-lab race

Over the past 12 months, competition among frontier models has converged into a close-range contest among three labs — Anthropic, OpenAI, and Google — though different leaderboards give different leaders. On Artificial Analysis's Intelligence Index, Claude Opus 5.5 ranks first with 58 points, while GPT-6 Astra and Google Gemini 4 Argon both score 53; on the user-vote-based Arena default leaderboard, Argon leads with 1,525 points, a group of Claude models follows at about 1,505 points, and OpenAI's best, GPT-5.6 Sol, ranks only 20th.

Chinese open-weight models overtook American ones in AI research papers in 2026