OpenRouter has shipped a proper usage analytics layer — a rebuilt Activity dashboard plus the new OpenRouter Analytics API — announced on 17 August 2026, and it means you can now see exactly which agent, which model and which API key is spending your money, then pull those same numbers programmatically from the terminal instead of squinting at a billing page. If you run AI agents that touch an OpenRouter key all day, this is the difference between guessing at your costs and actually managing them.
📺 Watch: Free Claude Code + Omniroute Is ABSURD!
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
I route a serious chunk of my agent workloads through OpenRouter, and until now the honest answer to "what did that workflow cost this week?" was a shrug and a scroll through the activity feed. So when this update landed I went through the announcement line by line. Here is what shipped, what the API actually exposes, and how I would use it to protect margins in an AI-powered business.
What the OpenRouter Analytics API Actually Shipped
According to OpenRouter's announcement, the update has two halves: a redesigned Activity dashboard for humans, and an Analytics API for scripts and agents.
The dashboard opens with five core numbers, each with its own sparkline so you can see movement at a glance:
- Total spend — what you have actually burned in the period.
- Request count — how many calls went out.
- Token volume — the raw fuel consumption behind those calls.
- Cache hit rate — how much of your prompt traffic is being served from cache instead of billed at full price.
- Blended cost per million tokens — the single most useful health metric, because it tells you what you are really paying across every model you touch.
Beyond the headline numbers, the announcement lists views for top users, top apps, spend by model, request volume by model, token breakdowns split into prompt, completion and reasoning tokens, prompt caching metrics, and a split between bring-your-own-key usage and OpenRouter credits. There is also an Explore view — a custom query builder — a Trends view sorted by movement, and saved charts you can keep private or share with everyone in your organisation.
Querying Your Usage From the Terminal
The half I care about more is the API, because dashboards are for checking and APIs are for automating. Per the announcement there are two endpoints. The first, a GET call to /api/v1/analytics/meta, returns the supported metrics, dimensions, filter operators and granularities — effectively the menu. The second, a POST to /api/v1/analytics/query, runs the actual query and returns aggregated data. Both require a management key rather than a normal inference key.
Queries can group results by up to two dimensions at once, and the dimension list is long: model, variant, provider, API key, app, user, workspace, origin, country, data region, finish reason, context length, session, generation and even custom user IDs. In practice that means questions like "spend by model for this one client's API key" or "token volume by provider, split by finish reason" become one request instead of a spreadsheet afternoon.
If you want the exact cost-control playbook I use across my own agents — dashboards, alerts and the routing setup that keeps my blended rate down — I walk members through it step by step inside AI Profit Boardroom → Get the cost-control playbook
Why Usage Analytics Is a Money Feature, Not an Admin Feature
The announcement includes a cookbook example that makes the business case better than any pitch: an agent-driven analysis spotted a preview model quietly consuming 6,200 dollars a month at roughly twenty-five times the organisation's blended rate, traced back to a single batch-pipeline key. One query, one finding, and a leak that would have swallowed most solo operators' entire AI budget was identified.
That story matches everything I have seen running agents in my own business. Cost problems are almost never spread evenly — they hide in one runaway workflow, one wrongly configured key, one model that seemed cheap until reasoning tokens entered the picture. This is exactly why I already obsess over reducing Claude Code token usage: the wins come from finding the one thing that is quietly on fire.
My own testing philosophy is the same on the quality side. I run every serious model through Goldie Bench, our own benchmark, precisely because vendor numbers never tell you how a model behaves on your real work — and cost numbers are no different. Blended cost per million tokens is to your wallet what a benchmark score is to quality: a single figure that makes comparison honest. Watch both and you stop paying frontier prices for commodity work.
📺 Watch: Ox Alpha Just Hit 16 TRILLION Tokens in 5 Days
How This Fits the Rest of the OpenRouter Stack
OpenRouter has been shipping at a serious pace, and the analytics layer slots neatly into the rest of it. If you are new to the platform, three pieces are worth understanding alongside this one.
First, the OpenRouter Fusion API — the panel-and-judge system that sends one prompt to several models in parallel and fuses the answers. Fusion is the feature that tempts you to run more models at once; analytics is the feature that tells you what that habit costs. The two belong together, and this page deliberately covers the money side while that one covers the quality side.
Second, the corporate backdrop: OpenRouter announced on 19 August 2026 that it is joining Stripe, which makes visibility into your dependency on the platform more relevant, not less. And third, the surface area keeps widening — the video generation API means video jobs will show up in the same spend data as your text calls, which is exactly when per-key tracking starts to matter.
For agent builders, the practical pairing is simple: give each agent its own API key, then group spend by key. That one-key-per-agent discipline is exactly how we built the Agent OS, our own agent operating system, and it is what makes every workflow auditable after the fact. I run Hermes with an OpenRouter key as standard — my Hermes agent OpenRouter guide covers that setup — and if you push routing further with Omniroute, the same discipline applies.
OpenRouter Analytics API: Quick Reference
| Piece | What it gives you |
|---|---|
| Activity dashboard | Five core metrics with sparklines: spend, requests, tokens, cache hit rate, blended cost per million |
| Explore | Custom query builder for slicing usage any way you need |
| Trends | Analytics sorted by movement, so spikes surface themselves |
| Saved charts | Keep views private or share with everyone in your organisation |
| GET /api/v1/analytics/meta | Lists supported metrics, dimensions, filter operators and granularities |
| POST /api/v1/analytics/query | Runs aggregated queries, grouped by up to two dimensions |
| Authentication | Management key, not a standard inference key |
OpenRouter Analytics API FAQs
What is the OpenRouter Analytics API?
It is the programmatic side of OpenRouter's usage analytics, announced on 17 August 2026. A meta endpoint describes the available metrics and dimensions, and a query endpoint returns aggregated usage data — spend, requests, tokens and more — grouped by up to two dimensions such as model, API key or app.
Do I need a special key to use it?
Yes. According to the announcement, the Analytics API authenticates with a management key rather than the normal API key you use for inference calls.
Can I see which of my agents is spending the most?
That is the headline use case. Give each agent or workflow its own API key, then group analytics queries by key or by app. The dashboard's top-apps and spend-by-model views answer the same question visually.
Does it track cached and reasoning tokens separately?
Per the announcement, token breakdowns split prompt, completion and reasoning tokens, and cache hit rate is one of the five headline metrics — so you can see both how much thinking you are paying for and how much caching is saving you.
Verdict: The Least Glamorous Feature That Will Save You the Most
Nobody gets excited about analytics until the first time it finds a six-grand leak. The OpenRouter Analytics API is the least flashy thing OpenRouter has shipped this month and probably the most profitable for anyone running agents at scale: real per-agent, per-model, per-key cost visibility, queryable from the same terminal your agents live in. If AI is part of how you make money, knowing your blended rate is not optional admin — it is margin management. Wire it up before your own batch-pipeline key writes the cautionary tale.
Want agents that earn more than they burn? Inside AI Profit Boardroom I share the full stack — Agent OS, routing, cost dashboards and the weekly calls where we fix each other's setups → Join me inside
