The AI agent platform claims over one trillion tokens processed in 90 days, with most usage driven by paying customers
Runable, a year-old AI agent platform that lets users build websites, slides, reports, videos, and audio through natural language prompts, has reportedly raised $21 million to scale its operations. The company says 60% to 70% of its more than one trillion tokens processed over the last 90 days came from paying customers, a ratio that most early-stage AI startups would love to claim.
To be clear: “tokens” here means large language model inference tokens, the chunks of text that AI models like Claude, GPT variants, and Gemini consume when generating outputs. Not the kind you trade on Uniswap.
From launch to $2M ARR in weeks #
Founded in 2025 by Umesh Kumar and Saksham Sarda, Runable is incorporated in Delaware but operates out of Bangalore, India. The company launched Runable 2.0 in early April 2026 and claims it hit $2 million in annual recurring revenue within roughly three weeks of that release.
That’s a steep trajectory for a company running with somewhere between 10 and 23 employees. The lean headcount means Runable is generating roughly $87K to $200K in ARR per employee, depending on the actual team size.
The user base sits at an estimated 700,000, built largely through Discord communities and organic channels rather than paid acquisition.
Users interact with Runable through web, Discord, and mobile interfaces, issuing natural language commands to generate creative and productivity outputs. The platform recently introduced AI Canvas and an iOS app, both aimed at deepening engagement and expanding the surface area of what agents can produce.
The token economics of AI agents #
The one-trillion-token figure over 90 days is worth unpacking. A single page of English text runs about 500 to 750 tokens. So one trillion tokens over three months translates to a staggering volume of AI-generated content, roughly equivalent to processing over a billion pages of text.
What makes Runable’s claim notable isn’t the raw volume but the composition. Having 60% to 70% of that usage come from paying customers suggests the platform isn’t just attracting tire-kickers. Free-tier users are typically responsible for the vast majority of compute at most AI startups, which means the company is burning GPU credits on people who will never convert.
The company runs these workloads inside cloud sandboxes, leveraging models from Anthropic, OpenAI, and Google.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our