Liner's $36.1M bet is that the footnote, not the model, is the enterprise product

The financing arithmetic says more than the positioning does. Liner had raised roughly $28.2M in its entire life before this round, so the Series C alone is about 1.28 times everything that came before ^[19], and it lands 22 months after the last raise, a gap the company spent adding agent products and starting a B2B motion ^[11]. Five of the nine named participants are new, and four of those five are a development bank, two securities firms and a private equity house ^[21][3]. That is late-stage institutional money underwriting a change in what gets sold, not more of what already worked.

What a corporate buyer procures here is not intelligence. It is the chain of custody: Liner sits between the prompt and the model that writes the reply, finding source material and grounding the output in something a reader can follow back ^[4]. There is no pretraining bill to defend and no weights to differentiate, which means the evaluation runs on whether the answer can be inspected. The company's own product split follows that logic, with separate lines for academic work, business writing and investment research ^[12].

HUMAIN ONE is the closest thing to evidence that the argument sells at enterprise prices, and the publisher reads it as a reference deployment precisely because data control, source quality and explainability are the buying criteria in that environment ^[10]. Liner says it won the engagement after being benchmarked against other global technology providers ^[9]. That is the company's account. No scorecard accompanies it, and the same is true of the Similarweb placement of 30th for AI-service web traffic that Liner reports for late July 2026 ^[14]. A vendor whose pitch is traceability is asking to be taken on trust for its two most persuasive numbers.

The benchmark leaves a residue worth naming. A 95.3 leaves 4.7 points unaccounted for, which is about one short fact-seeking question in 21 ^[20][6]. That residue is the product's justification rather than its embarrassment, since the reason to buy a citation layer is to catch the miss before it reaches a financial model or a policy decision ^[17]. It is also the reason the benchmark cannot close the sale on its own ^[7].

Consumer scale is distribution, not income. Fourteen million registered users across 220 countries ^[5] and four consecutive appearances in Andreessen Horowitz's consumer generative AI ranking ^[15] sit in a market where free substitutes cap what anyone will pay ^[16]. The enterprise tests Liner now inherits are integration reliability, privacy and support ^[16], none of which a retrieval score measures, and the new capital is going into R&D, infrastructure, global hiring and enterprise expansion in Korea and beyond ^[18].