Launching today
AgentBurn
See what AI agents really cost to run
1 follower
See what AI agents really cost to run
1 follower
AgentBurn tracks what production AI agents actually burn: tokens per task, cost per solved task and framework overhead. 30 agents, 22 priced models, 3 public benchmarks ā every field labelled source-backed, derived, or left empty when nothing is public. No invented numbers, no paid placements. Open dataset, leaderboards and a free 3-page sample.
Hey Product Hunt š
I built AgentBurn because every "agent cost" list I could find was either a price-per-million table or a number with no source.
So the rule for this dataset is simple: a field is printed only if a public source states it, or a documented formula derives it. Otherwise it stays empty and says "not published". That is why 20 of 30 tracked agents show no per-task metric ā and why the report prints a data-availability matrix rather than a made-up ranking.
What you can use today, free:
ā¢ /projects ā 30 agent profiles with per-field provenance and a source link on every number
ā¢ /leaderboard ā cost per solved task, framework overhead and public benchmark sheets
ā¢ /live ā telemetry from real model calls through the open SDK (measured, clearly labelled first-party, never dressed up as market data)
ā¢ a 3-page sample of the report, no email required
The paid bundle is $19.90: a 29-page PDF, the CSV + JSON dataset frozen at the edition snapshot, and an Excel calculator. One-time, no subscription.
Two things I would love feedback on:
1. Is "not published" the right default for missing data, or should the gap be displayed differently?
2. Which agents should I add next ā especially ones that publish their own token and outcome numbers.
Thanks for looking. I'll be in the comments all day.