<- all tokdocs

OpenAI Frontier Firms Data: The 6x Skills Gap and What 17.1x Means

By , founder of Mavric and Apso

View on TikTok ->

OpenAI's August 12, 2026 enterprise data shows that 19% of weekly active users at "frontier firms" use skills, against 3% at typical firms, a gap of about 6x. A TikTok from Robert Ta (@therobertta_) reports those figures accurately. It misreads the chart behind the "17.1x" token growth figure, and it credits OpenAI for an 8x spending stat that comes from an a16z chart.

What OpenAI's frontier firms data says about skills

The slide Ta shows is titled "Plugins and Skills adoption: top decile v. middle decile v. OAI" and credits "OAI Signals, 'What Frontier Firms are doing differently' Aug. 12, 2026." Its bars are:

  • Skills: 19% for frontier firms (top 10% of enterprises), 3% for typical firms (middle 10%)
  • Plugins: 21% for frontier firms, 9% for typical firms
  • OpenAI itself: 95% and 93%

The video says "OpenAI's own data says frontier firms use skills at 19% and typical firms use it at 3%. That's about 6x gap on that one capability." 19 divided by 3 is 6.3, so "about 6x" is fair. A RuntimeWire report of the same release says OpenAI published its enterprise studies on 12 August 2026 and that the top 10% of enterprise users adopt skills about six times as often as typical firms.

The y-axis on the slide reads "Share of weekly active enterprise users." The percentages describe how many active users at each group of firms use skills. They do not describe how many firms use skills.

Do skills make a firm a heavy AI user?

The data does not show that. The chart compares the top decile and middle decile of OpenAI's own enterprise customers, ranked by token use. Frontier firms are picked because they use the most tokens, so the chart shows that heavy users also use skills more. It does not show that adopting skills moves a firm into the top decile. The RuntimeWire report makes the same caution: the data comes from OpenAI's customers and does not show that these tools improve results.

What "17.1x since April 2025" really means

The video says "frontier firms grew output tokens 17.1x since April 2025 and typical firms 2.1x." The chart's y-axis reads "Output tokens (x Apr 2025 typical)," and its subtitle reads "indexed April 2025."

The base for both lines is the typical firm's April 2025 level. So 17.1x means frontier firms ended June 2026 at 17.1 times what a typical firm produced in April 2025. It does not mean frontier firms grew their own output 17.1 times. The frontier line starts above the typical line at April 2025, so its own growth is smaller than 17.1x. The typical firm's 2.1x is a true growth figure because its base is its own April level.

The chart labels the gap between them "8.3x." That is not exactly 17.1 divided by 2.1, which gives 8.14, and the frame does not show whether the label uses unrounded values or per-user averages.

How fast is the gap growing?

BankInfoSecurity's report of the OpenAI release defines frontier firms as the top 10% of enterprise customers ranked by output tokens per active user, with typical firms between the 45th and 55th percentiles. It says that in June, frontier firms produced 8.3 times as many output tokens per active user as typical firms, up from 2.6 times in January. That means the gap more than tripled in about five months, a fact the video does not use.

The same report notes that OpenAI treats token volume as an imperfect proxy for business value.

Where the 8x top 1% spend stat comes from

The video says "the top 1% of AI spenders outspend the top 10% by about 8x. And that's just totally off a bell curve." Its opening line credits everything to "OpenAI's own data."

The slide behind that figure is a different chart. It is the left panel of an a16z Growth deck page titled "Power Users Are Pulling Away," with a footer reading "© 2026 Andreessen Horowitz." The chart is titled "AI Power Usage: Top 1% Outspends The Top 10% By ~8X," and its y-axis reads "Median Panelist AI Vendor Spend." That describes panel data on spending across AI vendors, not OpenAI usage logs. The footnote naming the dataset is too small to read in the video, so the 8x figure could not be traced past the slide.

Is this "crossing the chasm"?

Ta says the gap is between "the innovators versus the everyone else right now" on Moore's technology adoption curve, and that "I don't even think we've reached the laggards of the market yet." Nothing on the slides places these groups on an adoption curve. Both groups are paying OpenAI enterprise customers, so even the middle-decile firm has already adopted the product. The innovator label is the speaker's framing.

He closes by recommending that teams treat skills as "first class artifacts in your code base" and says this will "optimize your token spend for value." No slide measures value per token or links skills to lower or better-spent tokens, and the clip shows no skill file or repository. The description links a longer YouTube video, which was not reviewed here.

Bottom line

The 19% versus 3% skills figures and the 8.3x token gap match OpenAI's August 2026 release as reported by RuntimeWire and BankInfoSecurity. The 17.1x figure is measured against the typical firm's April 2025 level, not frontier firms' own growth. The 8x top 1% spend chart is an a16z panel, not OpenAI data, and none of the sources show that skills cause better results.

Sources

Published November 5, 2026. Based on a TikTok by @therobertta_. Full video notes (markdown)