A clear scale.
An ambitious horizon.

What the numbers tell you.

The American Super Intelligence Index is a guide to what broadly available systems can do. We use Super Intelligence (SI), as the name for the wider field. That does not mean every model on the board is already superhuman.

A score records capability at a particular moment. The current leaderboard is the October 14, 2026 snapshot. Each entry includes a score, relative speed and cost tiers, a recorded date, and an entry status. Nationality adds no points.

The stage names belong to this publication’s framework. They are not scientific certification, a safety rating, or a promise that a model will suit every job. Think of the scale as a guide to the horizon, not a guarantee.

Six stages. No countdown.

General Intelligence (GI) names the broad-capability stages below. The Roman numerals I, II and III distinguish levels within this publication’s framework. The six stages describe increasingly broad, capable, and reliable systems. They are not a calendar. Equal spacing on the chart does not mean equal effort—or equal time—to reach the next stage. Ranges include their lower limit and stop before the next: 20–<40 means 20 to under 40. 100+ means 100 and above.

A worldwide field, an American perspective.

The active board covers systems broadly available through a product, application programming interface (API), platform, or another public channel on standard terms. Private internal models, invitation-only research previews, and systems that need specially restricted infrastructure sit outside it.

A system withdrawn from public availability belongs in the historical record rather than the active board. This snapshot records the board as of October 14, 2026; it does not promise that every entry is still available today.

We write about American leadership, but nationality is not a scoring factor. And a five-model snapshot is not a survey of every system in the world.

Terms & abbreviations.

A quick reference for the index and its coverage. Official company, product and program names stay intact; we do not invent expansions for brand names.

SI

Super Intelligence. Our name for the wider field; using it does not imply every system is already superhuman.

GI

General Intelligence. Broad capability across unfamiliar tasks and fields; GI-I, GI-II and GI-III are stage labels on this index’s scale.

SSI

Specialized Super Intelligence. A system that reliably surpasses the best human experts and institutions, using their best tools, across a substantial part of an important field.

GSI

General Super Intelligence. The 100-point threshold defined above, beyond the strongest coordinated human expert teams—not a capability claimed for today’s board.

AI

Artificial Intelligence. We retain this conventional abbreviation in official names, historical titles, quotations and identifiers; our own generic prose uses Super Intelligence.

API

Application programming interface: a way for software to use another system’s capabilities.

CPU / GPU

Central processing unit / graphics processing unit.

LLM

Large language model.

JSON

JavaScript Object Notation, the structured data format used for our public results.

PC

Personal computer.

FP4

Four-bit floating point, a low-precision number format.

GB / MB / TB

Gigabyte / megabyte / terabyte. GB/s means gigabytes per second; Gb/s means gigabits per second. In the health coverage, TB instead means tuberculosis.

USD / USA / U.S.

United States dollars / United States of America / United States.

AWS / AMD

Amazon Web Services / Advanced Micro Devices.

TSMC

Taiwan Semiconductor Manufacturing Company.

NSF / NASA

National Science Foundation / National Aeronautics and Space Administration.

HAI

Stanford Institute for Human-Centered AI, the source of the AI Index report. AI is expanded above.

NAIRR

National Artificial Intelligence Research Resource. We preserve the program’s official name.

OSS / OCP

Open-source software / Open Compute Project.

HSM

Hardware security module, a device that protects cryptographic keys.

NPP

NVIDIA 2D Image and Signal Processing Performance Primitives, a library for accelerated processing.

USFP

Ultra-Dense Small Form Factor Pluggable, the NVIDIA and Cisco optics form factor.

MONAI

Medical Open Network for AI, the medical-imaging project. AI is expanded above.

GALEX

Galaxy Evolution Explorer, the space telescope.

FIMS / SPEAR

Far-Ultraviolet Imaging Spectrograph / Spectroscopy of Plasma Evolution from Astrophysical Radiation, two names for the ultraviolet instrument discussed in our astronomy coverage.

DNA

Deoxyribonucleic acid.

RISC-V

The open instruction-set architecture, pronounced “risk-five.” RISC means reduced instruction set computer; V is the Roman numeral five.

VERID

VERified Infrastructure for Discovery Laboratories, the project name used in the cited research.

Names checked against Stanford HAI, NSF’s NAIRR page, NVIDIA’s NPP documentation, the USFP announcement, Project MONAI, NASA’s GALEX page, FIMS research, the SPEAR instrument paper and RISC-V International.

Which date means what?

Published is when an article or guide first appeared here. Updated identifies a dated revision; its note describes the scope. A presentation edit does not mean every historical claim was rechecked that day.

Benchmark snapshot is the date of the recorded model results—currently October 14, 2026. Publishing a new story or revising wording does not rerun the benchmark. Source documents may have their own earlier dates.

The news archive is newest first by original publication, not by later cosmetic edits.

Questions about the framework.