Sapiom raises $35M to route AI agents to cheaper models, with Anthropic as a backer

Sapiom, a San Francisco startup that sits between AI agents and the models they run on, has raised $35 million in a Series A round led by Dragonfly. The round brings total funding to $50 million, raised in just 11 months since the company was founded.

The investor list tells an interesting story on its own. Accel, Gradient, Coinbase Ventures, VanEck Ventures, Okta Ventures and Menlo Ventures all participated — but the most notable name on the cap table is Anthropic, one of the very model providers whose revenue Sapiom's core product is built to reduce.

The pitch is simple: make AI agents cheaper to run. When an agent performs a task, it typically calls a large language model, and most agent frameworks default to the most powerful — and most expensive — model available. Sapiom intercepts that decision and routes each call to the cheapest model that can still handle the task, based on complexity, cost, latency, reliability, and company policy.

Sapiom launched three products alongside the funding announcement: Router, which matches each model call to the most efficient option and is the core product; Agent Studio, which lets developers build, test and inspect agents locally within their existing codebase and deploy to production with a single command; and Runtime, which manages infrastructure for production agents, recording every model call, tool call and retry with cost and outcome in a unified control plane.

The platform has processed more than 270 million transactions since launching six months ago and now powers over 100,000 agent runs per day. One thing separates Sapiom from most routing competitors: it serves open-weight models from its own racks in a San Jose data center rather than acting purely as a middleman for API calls, charging for compute directly instead of adding a markup.

The clearest evidence of its value comes from Polsia, an AI startup that operates entirely through agent swarms with zero human employees. Polsia's projected revenue jumped from $100,000 to $10 million in a year, but its monthly spending on Anthropic's models climbed to $1.2 million; after Sapiom ran evaluations and optimized routing, that number dropped to roughly $100,000 per month — a reduction of nearly 10x.

Founder and CEO Ilan Zerbib frames the relationship with Anthropic as aligned rather than adversarial: he estimates 95% of agent tasks do not require a frontier model, and cheaper inference lets more companies deploy agents at scale, expanding the total market for models — including frontier ones for the tasks that genuinely need them.

The bigger picture supports the thesis: Gartner forecasts more than 40% of agentic AI projects will be canceled by the end of 2027, with escalating costs among the leading reasons, and a KPMG survey of 2,100 executives in June found only 7% could point to established returns from AI investments. The routing market itself has become crowded, with at least 80 active competitors; Amazon bundles routing into Bedrock, Microsoft builds it into Azure, and OpenRouter processes around 25 trillion tokens per week.

Sapiom also integrates with Coinbase's x402 protocol, an open standard designed to let AI agents interact directly with the real economy. Haseeb Qureshi, managing partner at lead investor Dragonfly, joins Sapiom's board; in his framing, AI agents are becoming employees without a manager or a budget, and CTOs are increasingly forced to act as CFOs, allocating real money with no visibility into where it goes.