The SMID Point

Research evidence excerpt

The SMID Point

Industry

Artificial Intelligence - Citi’s Inference Ahead - Constraints v. Capabilities

The realized risk from autonomous agents escaping sandboxes has moved from

interrupting lunch in April to infiltrating Hugging Face infrastructure in July. This

creates new fault lines around security and regulation while open-weight models

proliferate and grow more performant (Poolside, 7/21; Figure 2). Even as a swift

resolution to the debate (Nvidia, 7/24) on AI regulation (Scott Bessent, 7/22)

remains unlikely, the landscape raises the compliance bar for enterprises that

maintain their own abstraction layers for running models. Meanwhile, the

underlying interpretability gap persists, with the providers training these models

still unable to fully inspect why they act. As systems grow more autonomous and

the economic tradeoff between some agents and people shows potential to

narrow (METR, 7/21), token spend will continue to accelerate, driving the

increasing demand for infrastructure reflected in our updated AI Industry model.

Heath Terry | Shelby Spencer | Ashley Kim | Janna Withrow

A Summary of Signals

Metric Unit LatestDate Period Calculation Period-Over-Period Δ LatestPeriod 2ndPeriodLatest 3rdPeriodLatest 4thPeriodLatest 5thPeriodLatest 6thPeriodLatest Trendline

Proprietary and Open Models (Artificial Analysis)

Highest Intelligence for Proprietary Models Points 7/23/26 Week Point-in-Time — 0.0% 60 60 60 60 56 56

Highest Intelligence for Open Models Points 7/23/26 Week Point-in-Time — 0.0% 57 57 51 51 51 51

Gap in Proprietary and Open Intelligence Points 7/23/26 Week Point-in-Time — 0.0% 3 3 9 9 5 5

Highest Intelligence Models from the Top-20 Providers (Artificial Analysis)

Median Intelligence Points 7/23/26 Week Point-in-Time — 0.0% 43 43 40 39 34 34