Stanford HAI releases 2026 AI Index Report

Stanford’s Institute for Human-Centered AI published the 2026 edition of its annual AI Index, the report the field treats as its most comprehensive year-in-review of capability, investment and adoption. The report’s headline capability finding was speed: performance on SWE-bench Verified, a benchmark of real-world software-engineering tasks, “rose from 60% to near 100% in a single year.”

Alongside that, the Index highlighted what it called a “jagged frontier” in AI progress — models excelling unevenly rather than uniformly across tasks. Its illustrative example was stark: Google’s Gemini Deep Think won a gold medal at the International Mathematical Olympiad, yet the report found that the top model of the period read an analogue clock face correctly only 50.1% of the time, a task most children manage easily.

The rest of the report tracked the scale of the industry underpinning that progress: industry produced more than 90% of notable frontier models in 2025, organisational AI adoption reached 88%, US private AI investment reached $285.9 billion, and documented AI-related incidents rose to 362 from 233 the year before.

Coming from an academic centre with no commercial stake in the results, the Index is widely cited as a neutral benchmark of where the field stood at a given point. Its jagged-frontier framing became a recurring reference point through 2026 for arguments that benchmark scores alone overstate how generally capable frontier models are.