OpenRouter vs Sudowrite

Most of this window is developer guides — an image-generation tutorial for the Unified Image API, a vision request-body walkthrough, a tool-calling loop that swaps providers by changing one string, and two pieces on team spend controls. The exception is a real launch: usage analytics that break a team's spend down by agent, model and request, with saved charts, click-through from any bar into the logs behind it, and the same data queryable from a terminal through a new Analytics API. Live web search leaderboards grading engines, depth and models are the other release-shaped item.

OpenRouter has spent this year building the enforcement half of gateway governance — Guardrails, then team spend controls — and this closes the measurement half. Agent-level attribution is the new dimension: a team pointing a fleet of agents at one key could cap what they spent but not say which agent spent it. Shipping it with an API matters as much as the dashboard, because it means the gateway's aggregate usage data stops being only OpenRouter's routing moat and becomes something customers pull into their own systems.

With attribution in place, the plausible next move is acting on it — budgets or routing policies scoped per agent rather than per key. The benchmark surface should keep expanding alongside it, since both come from the same habit of measuring what actually happens rather than what is declared.