Cost per session replaces price per token as a model comparison
- Source
- x.com
- Date
New data layer in OpenRouter and our API: Cost per Session Cost per session is a way of assessing the real-world cost of using models, across tasks and harnesses. Here's how Kimi K3 actually compares to Sol, Fable, and Opus variants on Hermes:
It's more useful than price per token, which doesn't incorporate how token-efficient a model is. Cost per session encompasses all tasks across the market, as seen by each major harness, by looking at medians across 30 days. Here's what the Pareto frontier looks like with cost per session:
You can make your own charts like this by simply querying our Dataset API. The data is licensed under CC by 4.0. View the rankings here: https://t.co/z5HcJalLKD Or use the Dataset API: https://t.co/7oCDWKQPz4
Context
A low price per does not reveal how much an entire coding- session will cost, because models can consume different amounts of text while working. OpenRouter's Cost per Session metric summarizes observed spending using medians over 30 days. Its accompanying chart separates sessions by the agent software used and by turn-count bands, helping readers compare spending at roughly similar session lengths. The announcement therefore adds a usage-based view alongside unit prices. It measures session spending; it does not establish that differently priced sessions completed equally difficult work or achieved equally good results.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
Checking sign-in…
Loading comments…