Which cloud AI coding agents people actually run, by token volume on OpenRouter (usage data, not hype)

OpenRouter publishes usage rankings for cloud coding agents, ranked by actual tokens processed through the platform. It's one of the few adoption signals in this space that isn't stars, funding, or Twitter reach, it's what people are genuinely running at volume.

Current top of the cloud-agent category:

- Roo Code - 6.16B tokens

- Ito - 3.58B

- Letaido - 3.19B

- Agent Zero - 2.96B

- Clark - 2.17B

- goose - 1.82B

- TeleClaw - 861M

- Rayline - 801M

- GitLawb - 600M

What I find interesting about this list from an AI-industry angle:

- The names dominating usage are almost entirely different from the names dominating the conversation. No Cursor, no Copilot, no Claude Code in this specific category (they're mostly IDE/local, so it's not apples to apples), and instead a set of agents most people would struggle to name.

- Usage and mindshare are barely correlated. Roo Code processing 6B+ tokens while rarely surfacing in discussion says something about how noisy our sense of "what's winning" actually is.

- There's real architectural diversity here. Most are conventional agent harnesses, but the one at #9, GitLawb, is structurally different, a decentralized git network (repos on IPFS, signed commits, agents as first-class identities) where the coding agent is one component of a larger system. Seeing that clear 600M tokens alongside pure harnesses is a data point on whether agent-native infrastructure is finding real usage or just discourse.

Mostly sharing because I think usage data is underrated in how we evaluate this field. We tend to reason from launches and hype cycles when the actual behavior is measurable.

Anyone have insight into why the top few (Roo Code, Ito) pull the volume they do? Genuine adoption, or a handful of heavy automated users inflating it?

Source: openrouter.ai/apps/category/coding/cloud-agent