AgentMarginBlog
Blog
Notes on customer-level AI cost,
margins and what's shipping in AgentMargin.
Accuracy3 min read
Gemini 4 Argon doubles on December 1. Is your cost data ready?
Model prices now come with end dates. If your cost tracking keeps one price per model, you'll either rewrite history or undercount the new month. Price every call at the rate in effect when it ran.
Read post
Guides4 min read
How to track OpenAI API usage per user
OpenAI reports usage by project, API key and model. Your customers aren't any of those. Here's how to meter every completed response against the customer who caused it.
Read post
Concepts3 min read
LLM cost per user: why your average is lying to you
A healthy average AI cost per user can hide dozens of customers who cost more than they pay. Here's how to measure cost per user properly, and what to do about the long tail.
Read post
Guides3 min read
How much does it cost to run an AI agent? The math per run
An agent re-reads its whole context on every turn, so the tenth turn costs three times the first. A worked example with current prices, and what caching does to it.
Read post
Guides4 min read
Cached input tokens, explained: what they cost and how to count them
Prompt caching can cut input cost by 90% or more, but OpenAI and Anthropic count cached tokens differently. Get it wrong and your cost per customer is off before you've priced a single token.
Read post
Concepts2 min read
Know your margin, not just your tokens
Model dashboards tell you how many tokens you spent. They can't tell you which customer is profitable. That takes usage and revenue on the same customer.
Read post
Accuracy2 min read
A missing price is not a $0 expense
When AgentMargin can't price an event, it says so. Here's why unknown beats a confident zero, and what to do when a model shows up as unknown.
Read post
Release2 min read
AgentMargin v0.3.0-alpha: workspace pricing and a dark dashboard
Each workspace now keeps its own auditable, effective-dated model prices, and the local dashboard has been rebuilt in React.
Read post