Stanford HAI releases 2025 AI Index Report
The eighth annual report put US private AI investment at $109.1 billion in 2024, nearly twelve times China's $9.3 billion, and inference cost for GPT-3.5-level performance down over 280-fold since late 2022.
- Benchmarks & progress
- Ideas & essays
- Notable
Stanford’s Institute for Human-Centered AI published the eighth edition of its annual AI Index, a data-heavy survey drawing on benchmark results, investment figures, hardware trends and policy activity from the preceding year. Its central economic finding was a sharp fall in inference cost: the report estimated that running a system at GPT-3.5’s level of capability had become more than 280 times cheaper between November 2022 and October 2024, alongside roughly 30% annual declines in hardware cost and 40% annual gains in energy efficiency.
On capability, the report tracked rapid year-on-year gains on demanding benchmarks — double-digit percentage-point jumps on MMMU, GPQA and SWE-bench — and noted that the gap between open-weight and closed frontier models had narrowed sharply on some measures, from around 8 percentage points to under 2 within a year. It also reported that the performance spread between the best and tenth-best systems on a common benchmark had roughly halved over the same period, evidence that the field’s leading edge was becoming more crowded rather than dominated by a single lab.
On investment, the Index put 2024 US private AI funding at $109.1 billion, against $9.3 billion in China — a gap of roughly twelvefold — with generative AI alone drawing $33.9 billion globally, up close to 19% on 2023. The report has become one of the most widely cited annual reference points for the field’s trajectory, cited by journalists, policymakers and researchers throughout the following year as a baseline for how fast capability, cost and investment were moving.