Model

Anthropic prices a $1.4 trillion IPO as Databricks logs show its costliest model at 0.6% of enterprise tokens

Databricks' gateway logs show the costliest Claude tier at 0.6% of enterprise tokens and open-weight models near a third, hours before Anthropic prices a $1.4 trillion debut on premium-model revenue.

Richard TangRichard Tangthree-minute read
Share
Dario Amodei, third from left, standing with Japanese Prime Minister Sanae Takaichi and officials at the Prime Minister's Office in Tokyo
Dario Amodei, third from left, at the Prime Minister's Office in Tokyo on October 29, 2025, for a call on Prime Minister Sanae Takaichi.

Anthropic$350B — Anthropic, private, latest valuation $350B prices the biggest IPO in history tonight and trades Thursday, and the order book is six times oversubscribed. Hours before that price is fixed, the only large meter of what enterprises actually pay for models shows the company's flagship tier nearly unused and open-weight models at nearly a third of tokens. The revenue is real; the pricing power behind the valuation is what is being tested 135.

The one large meter shows the flagship tier routed around

Databricks$218.01B — Databricks, private, latest valuation $218.01B published its first enterprise model-usage report on November 16, drawn from Unity Gateway, the metering layer more than 10,000 enterprise customers use to reach several model vendors at once, with data through November 7 1. Two numbers carry it. Claude Fable 5.1, Anthropic's costliest model, fell from 5.5% to 0.6% of these customers' proprietary-model tokens in six weeks 1. Open-weight models rose from 7.3% to 30.9% of tokens in four months 1.

Read the limits first. The report gives no share for Fable 5.2, the model Anthropic released on November 9 and is selling now, no data on whether Fable 5.1 usage moved down to Opus 5.5 or out of the gateway, and no explanation of its July baseline; the closing note says the current window runs from September 21 to November 7 1. The measurement is blind exactly where the pricing question sits.

On Databricks' gateway, open weights neared a third of tokens while the flagship Claude tier sank to 0.6%

  • Claude Fable 5.1 (share of proprietary-model tokens)
  • Open-weight models (share of tokens)
0%10%20%30%40%July baselineNov window (21 Sep to 7 Nov)30.9%
Data
Claude Fable 5.1 (share of proprietary-model tokens)Open-weight models (share of tokens)
July baseline5.5%7.3%
Nov window (21 Sep to 7 Nov)0.6%30.9%
Percent of tokens, and of proprietary-model tokens for the Fable line, across Databricks Unity Gateway customers. The chart reads each measure against its own base: Fable 5.1's decline is measured over six weeks, the open-weight rise over four months. Databricks' November 16 report does not explain the July baseline. Source: SecondSource, November 17, 2026.1

The router's own demo cuts the bill

Databricks sells the control plane, and it is building the habit of routing models down. At its London event this month it demoed smart routing on Unity Gateway: simple tasks moved from frontier models to Claude Haiku 3.5, for a reported 12% spend reduction, about $150,000 a week, in a scenario where a fictional customer had burned $10 million of coding-agent budget in four months unnoticed 2. That is the mechanism: the gateway makes every token auditable, and once a CFO can see the bill, cheap models win the work that does not need frontier reasoning.

A trillion-four valuation rests on the premium tier holding

Anthropic's case is strong: $33.24 billion of revenue in nine months, a run-rate around $81 billion at the end of October, and a $26 billion private placement from the chipmakers and cloud giants 3. But the growth is decelerating and revenue per dollar of computing fell from 2.08 to 1.72 between the second and third quarters 5.

The gateway cannot yet say what Fable 5.2 wins

A valuation around $1.4 trillion is priced on the premium tier holding 5. The meter says enterprises are learning to route around it, and they hold the control plane to do it. The IPO price will be set by tonight's demand, including roughly $19 billion of expected index buying 5. What tells you whether the premium holds is the disclosure Databricks' gateway cannot yet make: what Fable 5.2 actually wins 12. If the next report still shows open weights climbing while the flagship tier idles, the application layer, not the model bill, is where the margin is migrating 12.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Reader comments

0 comments

    Model

    HeyGen's Avatar 6 scored 84% on realism, but its 42.2% pick rate was below chance

    Avatar 6 launched November 17, 2026, with an 84% realism figure and first place in a blind arena; the score doubles a below-chance pick rate, and the arena capped 97% of clips at 35 seconds. That is the evidence base under a company claiming $200 million of annual revenue.

    Model

    MiniMax hit a record low, about 90% below its January peak, as a US rule gates its rented Nvidia compute

    Washington's new remote-access rule puts a US license between China's listed model makers and the offshore Nvidia capacity they rent. MiniMax, the purest listed bet on Chinese foundation models, holds the price tag.

    Model

    OpenAI cuts flagship pricing 20% to match Anthropic as heavy users call GPT-6.2 Astra unusable

    OpenAI cut its flagship's price to match Claude Fable 5.2's rate, and the users who run models hardest rejected it, 24 hours before Anthropic priced a $1.4 trillion IPO on the assumption the quality gap holds. The independent labs say the gap is real, for now.

    Model

    Microsoft books Azure revenue on Chinese open models running on Baseten hardware it does not own

    Microsoft now bills customers' committed Azure dollars for open models from Chinese labs running on Baseten's hardware, a toll booth earning rent on compute it does not own.

    Model

    Alibaba stops giving away its best AI model, three days before earnings

    Alibaba launched Qwen4-Max on November 16 and kept the 2.4-trillion-parameter flagship behind its own paid API, promising open weights only for smaller models. Thursday's report is the first test of whether Qwen's three billion downloads become cloud revenue, and it lands two days after a US rule put the chips behind the model under license.

    Model

    Perplexity claims $1 billion of revenue on five times less spend than Anthropic's $1.4 trillion IPO

    Perplexity says it hit $1 billion in recurring revenue on a fraction of Anthropic's spend, three days before Anthropic prices a $1.4 trillion IPO on the opposite assumption. Whose economics hold will only be settled by a filing Srinivas now says could come in early 2027.

    Subscribe to TheInference for free