According to Stanford’s 2026 AI Index, artificial intelligence is sprinting while the rest of us are still lacing up. Despite predictions that development might hit a wall, AI top models continue to improve at an unprecedented rate.

The US and China are nearly neck-and-neck in model performance, with Chinese models like DeepSeek and local giants lagging only modestly behind the frontrunners from Silicon Valley. While the US has more powerful models and capital, China leads in research publications and robotics.

The benchmarks to measure AI progress are struggling to keep up as models quickly outpace their ceilings. Some are poorly constructed, while others can be gamed by overtraining on benchmark test data. Transparency is worsening, with companies like OpenAI, Anthropic, and Google no longer disclosing training details.

AI models have advanced to match human PhD-level understanding in science, math, and language but still struggle in areas like physical tasks and self-driving cars. Adoption rates are soaring: more than half of the world’s population now uses AI, faster than the personal computer or the internet.