LastAI cover
Sign in

RankingsMethodology

How the LastAI Index is calculated

The LastAI Index is a 0–100 score for how widely products and organisations in AI, quantum, space and robotics are used and known. It is calculated from public data anyone can check, and every value shows its source and measurement date. The map and official tiers use this index. The web, mobile and spending tabs show a16z’s original ranks.

Model version 1.3Latest snapshot 2026-11-12397 entries scored

Principles

  • Checkable data only

    Each entry shows the public observations used, such as traffic rank, token share and repository stars, with their source and date.

  • Thin evidence, lower score

    Entries with little measurable data are scored down accordingly and labelled with an evidence grade.

  • Member opinion is separate

    Votes and star ratings carry zero weight in AI popularity scores. Select Member ratings to see the community’s view.

  • No ads, affiliates or share prices in the score

    Affiliate links, advertising, share prices and market caps are not inputs to the formula at all, and we do not edit ranks by hand.

What goes into AI popularity?

For products, a16z monthly web traffic and mobile active-user ranks share a weight of 40: 40 for a single observed channel or 20 each for two. Tranco domain rank is the fallback when neither is current. US card spending rank does not enter the index. For model makers we give the most weight to their weekly token share on OpenRouter. The numbers below are LastAI’s chosen weights, not market shares.

Products, services and technologies

  • Consumer web traffic rank

    Weight 20 or 40

    Web rank in a16z's Top 100 Gen AI Consumer Apps (Similarweb monthly visits). Only services on the list.

    Scale · 100 × (1 − log10(rank) ÷ 3). Rank 1 scores 100, rank 10 scores 67, rank 100 scores 33.

    Source · a16z Top 100 Gen AI Consumer Apps · values older than 365 days are left out

  • Mobile active-user rank

    Weight 20 or 40

    Mobile rank in a16z's Top 100 Gen AI Consumer Apps, based on Sensor Tower monthly active users, not downloads.

    Scale · 100 × (1 − log10(rank) ÷ 3). Rank 1 scores 100, rank 10 scores 67, rank 100 scores 33.

    Source · a16z Top 100 Gen AI Consumer Apps · values older than 365 days are left out

  • Web reach

    Weight 35

    The product domain’s Tranco rank (geometric mean of up to 30 recent ranks). It measures relative domain popularity, not a visitor count.

    Scale · 100 × (1 − log10(rank) ÷ 6). Rank 1 scores 100, rank 1,000 scores 50, outside the top million scores 0.

    Source · Tranco · values older than 10 days are left out

  • Developer interest

    Weight 20

    GitHub stars on the public repository. Only entries with a repository.

    Scale · 100 × (log10(stars) − 2) ÷ 3.6. 100 stars or fewer score 0, 10,000 score 56, 400,000 or more score 100.

    Source · GitHub REST API · values older than 10 days are left out

  • Revenue market share

    Weight 20

    The product's own revenue market share, with a verified source, period and comparison market.

    Scale · 100 × √(share ÷ 100). A 25% share scores 50.

    Source · 회사 공식 자료 · values older than 548 days are left out

Companies and organisations

  • Real API usage

    Weight 40

    The maker’s observed share of all tokens on OpenRouter over seven days. Only tokens from models in each day’s top 50 are attributed to a maker.

    Scale · 100 × √(share ÷ leader's share). The leader scores 100; a quarter of the leader's share scores 50.

    Source · OpenRouter · values older than 10 days are left out

  • Flagship product

    Weight 25

    The highest LastAI Index among the organisation's products in the same field. Products without enough evidence are not counted.

    Scale · The product's LastAI Index is used as is.

    Source · LastAI 지수

  • Developer interest

    Weight 10

    GitHub stars on the public repository. Only entries with a repository.

    Scale · 100 × (log10(stars) − 2) ÷ 3.6. 100 stars or fewer score 0, 10,000 score 56, 400,000 or more score 100.

    Source · GitHub REST API · values older than 10 days are left out

Weights for quantum, space and robotics

Products · Web reach 35 · Public attention 25 · Consumer web traffic rank 15 · Developer interest 20 · Revenue market share 20 · Member ratings 0–10

Organisations · Public attention 30 · Developer interest 10 · Real API usage 10 · Flagship product 25 · Member ratings 0–10

In these three fields public attention uses English-language article views, and member weight grows with votes and ratings.

How the score is put together

  1. Each value becomes 0–100 points

    Ranks and stars use logarithmic scales; shares use square roots. Each scale and its reference points are shown above.

  2. Take the weighted average

    Add each score times its weight, then divide by the sum of included weights. Missing and stale data are left out. A confirmed absence from the Tranco list counts as an observed zero.

  3. Reduce for thin evidence

    If the weights of the measured values add up to less than 40, the average is multiplied by that ratio. With a total weight of 20, the score is half the average (20 ÷ 40).

Worked example

For example, a product with no traffic rank and only a Tranco domain rank of 1,000 starts at 50. With just 35 in evidence weight, the final score is 50 × 35 ÷ 40 = 43.75. This is neither a user count nor a market share.

Organisations receive the highest index among their products as the 'flagship product' value. Products never inherit data from their parent company.

Evidence grades and tiers

The evidence grade describes data coverage. It is not a statistical confidence level or margin of error. It depends on the total weight of the measured values.

  • Strong evidencetotal weight 55 or more35 entries
  • Fair evidencetotal weight 40 or more86 entries
  • Thin evidencetotal weight 15 or more246 entries
  • Not enough evidencetotal weight under 1530 entries

Tiers (S to D) rank scored entries within the same field and kind (products with products, organisations with organisations). The top 8% are S, up to 23% A, up to 48% B, up to 75% C, and the rest D. Entries scoring 0 are D, and entries graded 'not enough evidence' cannot be S or A. Votes break ties between equal scores.

No tier is currently fixed by hand.

When it is recalculated

  • Index, ranks and tiers: recalculated every 15 minutes.
  • Domain rank is checked daily. GitHub stars every 6 hours, API usage once a day.
  • Daily snapshot: the first calculation after 00:00 UTC is stored once and never changed afterwards.
  • Weekly report: every Monday's snapshot is compared with the one a week earlier (same formula only) to show rises, falls and new entries. See the weekly report

Known limits

  • Web reach is measured per domain. Products that live under a large company's address, such as gemini.google.com, cannot be measured separately, so this value is left out for them.
  • GitHub stars show interest in a repository; they count neither installs nor active developers.
  • Real API usage covers traffic through OpenRouter only, not direct use of each company's own API.
  • The seventh a16z edition compares August 2026 web traffic (Similarweb) and mobile active users (Sensor Tower) separately. Ordinal ranks cannot tell us user totals or the size of the gap between products. Being unlisted does not mean having no users.
  • The index measures how widely something is used and known. It does not measure quality, safety or investment value.

Cite and download

The scores, ranks and tiers LastAI calculates may be used anywhere, commercially too, with attribution (LastAI Index, snapshot date and URL) under CC BY 4.0. Values at a dated URL never change, so please cite the dated URL. Data use termsBadges and tables

Suggested citation

LastAI (2026). LastAI Index v1.3, snapshot of 2026-11-12. https://lastai.org/rankings/index.json?date=2026-11-12

Download table (CSV, 2026-11-12)Data for programs (JSON)

For another date, change the date at the end of the URL (for example /rankings/index.csv?date=2026-11-10). Snapshots start on 2026-11-10. Outside values in the data (such as domain and web traffic ranks) belong to their sources, and each source URL is included.

Report an error

If a value is wrong, a source or official site is linked incorrectly, or a product or company is missing, please tell us. We check and fix it, and any change to the formula is recorded in the change log below. Contact details are on the About LastAI page.

Change log

  • Version 1.3 ·

    The AI index now gives more weight to product web traffic and mobile active-user ranks and model makers’ API usage. Article views and member ratings no longer count towards AI scores. Other fields and earlier snapshots are unchanged.

  • Version 1.2 ·

    Market value (market capitalisation of listed companies) is no longer part of the index. LastAI no longer collects or shows share prices or market caps. Prices never entered the calculation, so no score changed.

  • Version 1.1 ·

    Market value now uses an absolute scale on market capitalisation. The old scale gave even tiny listed companies about 45 points and added 35 when a company was the only listed one in its sub-field. No score changed, because market prices are not flowing yet.

  • Version 1.0 ·

    First public version. Scores now rest on outside data (web reach, Wikipedia pageviews, real API usage, GitHub), and member ratings count only as participation grows.

Real usage rankings of AI models and apps are calculated separately. See the AI usage methodology

You

My accountPast forecastsMy wallet