How to track AI spend across OpenAI, Anthropic, OpenRouter and xAI

If you build with AI in 2026, you probably don't have one AI bill. You have an OpenAI bill, an Anthropic bill, an OpenRouter balance quietly draining, and maybe an xAI experiment you forgot about. Each provider reports usage in its own console, its own format, its own units. Getting a single honest number for "what does AI cost me" is unreasonably annoying — here's the map.

Where each provider hides the number

Anthropic

The Console has a usage and cost view per workspace, and organizations get an Admin API that can report usage and cost programmatically. Watch out for the workspace split: if your team uses several workspaces, no single chart shows the total. Claude Code usage billed through an API key shows up here too — subscription usage (Pro/Max) doesn't, which is why Claude Code's local stats are worth reading separately.

OpenAI

The platform dashboard has a usage page with cost breakdowns by project and API key, plus a usage API for pulling the data out. The classic gotcha is project sprawl: costs are split per project, and the default view may not be showing you all of them.

OpenRouter

OpenRouter is prepaid credits, so the number you feel is the balance going down. The activity view breaks spend down per model and per request — in some ways the most transparent of the bunch, because routing through one gateway means one ledger across dozens of models.

xAI

The xAI console reports usage and billing per team. It's the youngest dashboard of the four and the one most likely to have moved things around since you last looked.

Why this stays annoying

• No shared format. One provider reports dollars, another tokens, another credits. Cache tokens, batch discounts and reasoning tokens are counted differently everywhere.
• Invoices lag. The dashboard says one thing mid-month; the invoice says another. Usage views are near-real-time, billing is not.
• Nobody sums it. No provider has any incentive to show you what you spend elsewhere. The total is your problem.

Getting to one number

The pragmatic approach: pick one unit (tokens, plus a notional USD value at list prices), pull every provider into it, and look at one chart. You can build that yourself with the usage APIs and a spreadsheet — a fine weekend project, the kind that itself burns a few million tokens in agent calls.

Or let Burnmaxxing do it: connect Anthropic, OpenAI, OpenRouter or xAI, and your verified burn is fetched, summed and ranked on the global leaderboard. By default your API key is used for exactly one read and then discarded — verification without storage. You get the one number, a percentile to go with it, and a cardto make it everyone else's problem.