AI Ranking Observatory

See who leads AI

27 charts · 540 entries · Continuously updated

A complete set of charts across AI websites, mobile apps, browser extensions, model benchmarks and Claude skills. Continuously updated.

As of September 2026, ChatGPT leads the global AI website traffic chart with 5.9B monthly visits.

ACT 01

Where People Meet AI

The first layer is the entry point. The same users arrive by three paths — a website, an app, a browser extension — and the three leaderboards do not line up. The winner on the web is not always the winner on mobile, and the most-installed extension is rarely the best-reviewed one.

On web and mobile the top spot is held by the same product, ChatGPT; on the browser-extension path the most-installed is Grammarly (40.0M installs).

Grammarly is rated 4.5; among the 52 candidates that clear a rating-and-review-count floor, the leader is Sider (rated 4.9, 5.0M installs).

ACT 02

The Map Is Not Flat

A global chart is a sum, and summing flattens structure. One product can be several times stronger in one country than another — and a product’s centre of gravity on the web is often not where it is on mobile. This act takes the totals apart.

ChatGPT draws its largest share of web visits from the United States (21.1%), while its mobile downloads concentrate most in India (33.3%).

Centre of Gravity

Key Markets

ACT 03

Where the Wind Blows

A stock chart may not change in a year; a growth chart changes every month. It answers a different question — where attention is moving. The cost: most of these names are unfamiliar, and high growth starts from a low base, which is why every growth chart carries a volume floor.

The fastest-growing AI website this period is Experiential Labs, up +40559% month over month, on 2.3M visits.

ACT 04

Two Rulers

How strong a model is depends on who scores it. One ruler is machine evaluation on standard test sets, the other is human votes in anonymous head-to-head matches. They do not measure the same thing, so this act keeps them side by side and leaves the disagreement visible.

The machine-evaluation leader is Anthropic’s Claude Fable 5.1 (intelligence index 61.3); the blind human-preference leader is Anthropic’s Claude Fable 5 (1506 Elo). Snapshots taken October 1, 2026 and September 15, 2026 respectively.

ACT 05

Who Actually Does the Work

Benchmarks decide what gets discussed; call volume decides what does the work. One chart only, but it is the control group for the act before it: on a multi-model gateway, where the tokens actually go — a list that often differs from the benchmark leaders.

The benchmark leader is Anthropic’s Claude Fable 5.1, while the most-called model on the aggregation gateway is Z.ai’s GLM 5.3 Flash, consuming 58.6T tokens this period. The two are measured at different points: October 1, 2026 and September 28, 2026.

ACT 06

Who Builds the Layer Above

The top layer is not owned by the model vendors. As base capability converges, the difference shows up elsewhere: someone packages it into a reusable skill for others to use. This act is about those people — what they publish, and who keeps shipping.

Downloads are led by 腾讯文档 TENCENT DOCS (1.3M), stars by self-improving agent (4.0K); the most prolific creator is Iván, with 834 skills published.

01

Methodology

Let the data speak: every ranking is decided by its metric alone, on one consistent and comparable basis, with no editorial opinion and no commercial influence.

02

Data Sources

Website traffic, mobile app downloads, browser extension installs, model benchmark leaderboards and API usage, plus the Claude skills ecosystem.

03

Last Updated

Data is updated continuously and every chart shows the latest release from its source; each card carries the data period it covers.